A vehicle side collision warning system based on information fusion

Abstract

A vehicle side collision warning system based on information fusion is disclosed, comprising a signal acquisition module, a vehicle state estimation module, a vehicle trajectory prediction module, and a vehicle collision warning module. The signal acquisition module detects the relative position of the target vehicle relative to the vehicle itself and outputs it as a measurement variable to the vehicle state estimation module. The vehicle trajectory prediction module estimates the relative motion trajectory of the vehicle and sends it as input to the vehicle collision warning module. The vehicle collision warning module calculates the critical value of the square root of the sum of the squares of the lateral and longitudinal distances of the target vehicle relative to the vehicle itself within the safe collision warning time, thereby determining whether there is a risk of side collision.

Description

Technical Field

[0001] This invention belongs to the field of active safety in automobiles, and relates to a vehicle collision warning system, and more particularly to a vehicle side collision warning system based on information fusion. Background Technology

[0002] With the improvement of people's living standards and the acceleration of the pace of life, automobiles, as a convenient means of transportation, have become the first choice for people's travel. Road traffic accidents have also gradually become a serious threat to people's lives and property. Among them, two-vehicle and multi-vehicle collisions are one of the most serious types of road traffic accidents. Some ramp merging and intersections without traffic lights are extremely prone to vehicle collisions. Once a side collision occurs, it often leads to serious casualties and significant economic losses. Therefore, issuing warnings before a car collision occurs is an important measure for vehicle safety and driver assistance to avoid collision accidents. Vehicle collision warning systems mainly have two warning modes: one is based on a safe collision distance model with reference to real-time vehicle speed and acceleration; the other is based on predicting the collision time according to the vehicle's trajectory and referring to a safe collision time model. Both methods require prediction of the vehicle's driving state and trajectory; therefore, accurately estimating the state and predicting the trajectory are key technologies for achieving collision warning.

[0003] Existing research often employs Kalman filters to estimate vehicle states based on vehicle dynamics or kinematic models. For example, unscented Kalman filters have been used to compare the estimation accuracy of different vehicle kinematic models, demonstrating that constant turning rate and speed models, as well as constant turning rate and acceleration models, can achieve good vehicle motion state estimation accuracy in urban and highway environments. Another example is the development of a seventh-order capsular Kalman filter algorithm by extending spherical surface integrals and radial integrals to the standard capsular Kalman filter algorithm, improving estimation accuracy and robustness while reducing computational cost. Furthermore, a square root factor has been introduced into the standard capsular Kalman filter to propose a square root capsular Kalman filter, ensuring the positive definiteness and symmetry of the error covariance matrix and effectively enhancing the robustness of the estimated values. However, in these methods, the estimation accuracy still needs improvement due to the limited degrees of freedom in the chosen state transition equations and the lack of consideration for the time-varying nature of vehicle speed.

[0004] In vehicle trajectory prediction, current main methods include trajectory prediction based on intent recognition, trajectory prediction based on driving behavior, and trajectory prediction based on physical motion models. For example, one method obtains the target vehicle's state information based on driving behavior and high-precision map information, and predicts the vehicle trajectory using lane curvature constraints. This method is highly dependent on the measurement accuracy of sensors and requires a large amount of historical state information for training. Another method uses a constant turning rate and acceleration model to predict future vehicle trajectories through state iteration. However, the accuracy of this motion model-based trajectory prediction is overly dependent on the accuracy of the prediction model. Once the actual vehicle operating conditions do not match the motion model, i.e., when there is uncertainty in the operating conditions, the trajectory prediction accuracy is difficult to guarantee. Therefore, this method is not adaptable to complex operating conditions and cannot handle vehicle trajectory prediction under complex driving behaviors. In side collision warning, collision warning algorithms calculate the risk of collision, mainly using safe time algorithms and safe distance algorithms. Among these, the safe time algorithm has proposed the concept of "immediate collision time," but it does not consider the driver's factors.

[0005] In practical use, existing vehicle side collision warning methods have the following shortcomings:

[0006] 1. Existing technologies are inadequate for predicting driving trajectories using vehicle motion models when performing collision warnings for vehicles with complex driving behaviors.

[0007] 2. Existing technologies do not consider the influence of the road surface adhesion coefficient on the maximum deceleration during continuous braking time, and the safety collision warning time cannot accurately assess the collision risk.

[0008] 3. Existing technologies do not correct for relative vehicle speed when calculating the safety collision warning time, making it impossible to accurately assess collision risk in side collision scenarios.

[0009] 4. The existing technology uses a state transition equation with few degrees of freedom and does not consider the time-varying nature of vehicle speed, so the estimation accuracy still needs to be improved.

[0010] 5. Existing technologies do not propose an early warning method that integrates information fusion using grey model prediction and Kalman filtering. When the vehicle state estimator performs multi-step relative position prediction, it cannot perform measurement correction due to the lack of measurement variables for future time moments, and the accuracy of vehicle state estimation cannot be guaranteed. Summary of the Invention

[0011] To overcome the above problems, the present invention proposes a solution that addresses multiple problems simultaneously.

[0012] The technical solution adopted by this invention to solve its technical problem is: a vehicle side collision warning system based on information fusion, wherein the warning system includes a signal acquisition module, a vehicle state estimation module, a vehicle trajectory prediction module, and a vehicle collision warning module. The signal acquisition module detects the relative position of the target vehicle relative to the vehicle itself and outputs it as a measurement variable to a first vehicle state estimator. The vehicle state estimation module, which is the first vehicle state estimator, is used to perform continuous m-step state estimation of the relative state of the target vehicle relative to the vehicle itself based on the acquired relative measurement vectors, serving as the raw data for a gray model predictor. The vehicle trajectory prediction module includes a gray model predictor and a second vehicle state estimator. The gray model predictor performs n-step measurement prediction using GM(1,1) on the most recent m sets of measurement vectors, and sequentially inputs the obtained k+1 to k+n step measurement prediction values ​​as the actual measurement vectors of the second vehicle state estimator. The second vehicle state estimator estimates the relative motion trajectory of the vehicles, serving as input to the vehicle collision warning module. The state variables of both the first and second vehicle state estimators include the lateral distance, longitudinal distance, relative speed, relative yaw angle, relative yaw rate, and relative acceleration of the target vehicle relative to the vehicle itself. The measurement variables of both the first and second vehicle state estimators include the relative distance between the target vehicle and the vehicle itself, the rate of change of relative distance, and the azimuth angle. The state transition equations for both the first and second vehicle state estimators are constant turning rate and acceleration models.

[0013] The vehicle collision warning module calculates the safe collision warning time t. min,k Within the range, the square root of the sum of the squares of the lateral and longitudinal distances of the target vehicle relative to the current vehicle is checked against a critical value. If the value is less than the critical value, a side collision risk is identified, and a collision warning is issued to the current vehicle. If the value is greater than the critical value, no side collision risk is identified, and the next warning cycle begins.

[0014] The vehicle collision warning module calculates the safe collision warning time t. min,k This includes driver reaction time, brake coordination time, deceleration increase time, and continuous braking time, corrected for by relative yaw angle, and expressed as...

[0015]

[0016] Where k is the discrete time step, t min,k Let t1 be the safety collision warning time at the k-th warning step, t2 be the driver's reaction time and the brake coordination time, t3 be the acceleration increase time, and v be the acceleration increase time. x Let μ be the vehicle's longitudinal velocity, μ be the road surface adhesion coefficient, and g be the acceleration due to gravity. t4 is the relative yaw angle, and t4 is the continuous braking time, used to account for the influence of the road surface adhesion coefficient on the maximum deceleration. Before the relative yaw angle correction, t4 is given as v. x / (μg)-0.5t3.

[0017] Preferably, the safe collision warning time calculated by the vehicle side collision warning method includes driver reaction time, brake coordination time, deceleration increase time, and continuous braking time.

[0018] Preferably, when the vehicle side collision warning method performs multi-step relative position prediction, after the signal acquisition module acquires the relative motion state of the target vehicle, it first performs m-step state estimation independently through the first vehicle state estimator to realize the relative motion state estimation of the vehicle.

[0019] Then, starting from step m, we enter the vehicle trajectory prediction part; using m sets of measurement vectors in time k-m+1, k-m+2, ..., k, we perform gray model prediction, with n prediction steps, to obtain the predicted values ​​of the measurement vectors at time k+1, k+2, ..., k+n, which are used as the actual measurement vectors of the second vehicle state estimator.

[0020] In the first step (k+1) trajectory prediction, the second vehicle state estimator uses the vehicle relative motion state estimates from the k steps. As the initial state, the grey model prediction value of the measurement vector As the measured vector, the relative motion state of the vehicle at step k+1 is estimated.

[0021] The second vehicle state estimator uses the vehicle relative motion state estimates from k+1 steps. As the initial state, the grey model prediction value of the measurement vector Using the measured vectors, the relative motion state of the vehicle at step k+2 is estimated.

[0022] By analogy, the second vehicle state estimator eventually estimates the relative motion state of the vehicle over k+n steps.

[0023] Preferably, in the vehicle side collision warning method, when issuing a collision warning, the second vehicle state estimator predicts the relative motion state of the vehicle at time k+n in the future. Used to determine whether a collision will occur.

[0024] Preferably, when the vehicle side collision warning method issues a collision warning, if it determines at time k that a collision will not occur, it proceeds to the next step of trajectory prediction and performs measurement variable correction using the following method: first, it performs state estimation at time k+1; then it predicts the vehicle trajectory at time k+n+1; during this process, it uses a second vehicle state estimator to correct the measurement variables of the gray model predictor.

[0025] In summary, this invention improves state estimation and trajectory prediction in two aspects: For vehicle state estimation, based on the easily measurable motion state of the target vehicle relative to the vehicle itself, a method combining a constant turning rate and acceleration model with square root decimal Kalman filtering is proposed, improving the accuracy of motion state estimation. For vehicle trajectory prediction, a relative motion trajectory prediction method incorporating gray model prediction and square root decimal Kalman filtering is proposed, improving multi-step prediction accuracy. Based on the relative motion trajectory prediction results, the collision time is calculated based on the predicted collision position, ultimately establishing a vehicle collision warning method.

[0026] The beneficial effects of this invention are:

[0027] 1. Regarding point 1 in the background technology, the present invention improves the accuracy of collision warning by fusing information through Kalman filtering and grey model prediction.

[0028] 2. Regarding point 2 in the background art, the present invention considers the influence of the road surface adhesion coefficient on the maximum deceleration by establishing the relationship between the continuous braking time and the road surface adhesion coefficient, thereby improving the effectiveness of the safety collision warning time assessment of collision risk.

[0029] 3. Regarding point 3 in the background art, the present invention uses the cosine of the relative heading angle to correct the relative vehicle speed, thereby improving the effectiveness of the safety collision warning time assessment of the side collision risk.

[0030] 4. Regarding point 4 in the background technology, this invention selects a high-dimensional vehicle motion model as the state transition equation, and considers the change in vehicle speed through acceleration to improve the accuracy of vehicle state estimation and further improve the accuracy of multi-step relative position prediction.

[0031] 5. Regarding point 5 in the background technology, this invention makes improvements in three aspects: firstly, it proposes an early warning method that integrates information fusion using gray model prediction and Kalman filtering; secondly, it proposes a vehicle side collision judgment criterion; and finally, it establishes a measurement variable correction method. These methods can improve the accuracy of vehicle side collision early warning.

[0032] Note: The above improvements are not listed in any particular order, and each one makes the present invention different from the prior art and represents a significant advancement. Attached Figure Description

[0033] Figure 1 This is a diagram showing the components of the vehicle side collision warning system of the present invention;

[0034] Figure 2 This is a flowchart of the vehicle trajectory prediction process of the present invention;

[0035] Figure 3 This is a flowchart of the vehicle collision warning algorithm of the present invention;

[0036] Figure 4 This is a diagram of the relative motion model of the target vehicle;

[0037] Figure 5 It is a diagram of the vehicle braking process;

[0038] Figure 6 This is a schematic diagram of a side collision;

[0039] Figure 7 It refers to the safe collision warning time under different road surface adhesion coefficients;

[0040] Figure 8 Collision warning under different road surface adhesion coefficients. Detailed Implementation

[0041] A vehicle side collision warning system based on information fusion, the composition of which is as follows: Figure 1 As shown, it includes a signal acquisition module 20, a vehicle state estimation module 30, a vehicle trajectory prediction module 40, and a vehicle collision warning module 50.

[0042] Vehicle state estimation module 30 is the first vehicle state estimator 31;

[0043] The vehicle trajectory prediction module 40 includes a gray model predictor 41 and a second vehicle state estimator 42;

[0044] The signal acquisition module 20 detects the relative position of the target vehicle 10 with respect to the vehicle 60 and outputs it as a measurement variable to the first vehicle state estimator 31;

[0045] The first vehicle state estimator 31 estimates the relative motion state of the vehicle, which serves as the raw data for the gray model predictor 41.

[0046] The gray model predictor 41 performs relative position prediction, which serves as the input to the second vehicle state estimator 42;

[0047] The second vehicle state estimator 42 estimates the relative motion trajectory of the vehicle and uses it as input to the vehicle collision warning module 50.

[0048] The vehicle collision warning module 50 calculates whether there is a risk of side collision within the safe collision warning time; if so, it issues a collision warning to the vehicle 60.

[0049] If the square root of the sum of the squares of the lateral and longitudinal distances of the target vehicle 10 relative to the vehicle 60 is less than a critical value, it is determined that there is a risk of side collision and a collision warning is issued to the vehicle 60; if it is greater than the critical value, it is determined that there is no risk of side collision and the next warning cycle begins.

[0050] Furthermore, as a vehicle side collision warning method, the state transition equations of the first vehicle state estimator and the second vehicle state estimator are the same, and a constant turning rate and acceleration model are adopted.

[0051] The vehicle state estimation method based on Kalman filtering first predicts the state using the vehicle state transition equation, and then corrects the predicted state based on the actual measurement signals. Therefore, the vehicle state transition equation needs to reflect the complexity of the actual vehicle motion. Considering the relationship between the vehicle state vector and the measurement vector, this invention selects the lateral distance, longitudinal distance, relative speed, relative yaw angle, relative yaw rate, and relative acceleration between the target vehicle and the vehicle as state estimation quantities. The relative motion of the two vehicles is simplified to the absolute motion of one vehicle, such as... Figure 4 The constant turning rate and acceleration model shown has a constant yaw rate and acceleration during the driving process.

[0052] The vehicle state transition equation is:

[0053] x k =f(x) k-1 )+ω k (1)

[0054] In the formula, Let p be the vehicle state vector, where p k q represents the longitudinal distance between the target vehicle and the current vehicle. k v represents the lateral distance between the target vehicle and the current vehicle. k The relative velocity is defined as the vector difference between the velocity of the target vehicle and the vehicle itself, and is taken as a positive value. The relative yaw angle is defined as the angle between the relative speed of the target vehicle and the vehicle itself and the X-axis, with the counterclockwise direction being positive. The relative yaw rate between the target vehicle and the vehicle itself; a k ω is the relative acceleration; k This is process noise.

[0055] When the relative yaw rate is zero, the constant turning rate and acceleration model cannot predict lateral and longitudinal distances. Therefore, the vehicle state transition equations need to consider both cases where the relative yaw rate is zero and not zero. When the relative yaw rate is not zero, the constant turning rate and acceleration model is as follows:

[0056]

[0057] When the relative yaw rate is zero, the constant turning rate and acceleration model is:

[0058]

[0059] In the formula, T is the calculation step size.

[0060] The constant turning rate and acceleration model assumes constant yaw rate and acceleration, but this assumption is difficult to maintain in real-world driving. Therefore, process noise is mainly generated by disturbances in relative yaw acceleration and the rate of change of relative acceleration. Specifically, the relative acceleration noise ω... a It can be obtained by integrating the rate of change of random relative acceleration over time, and can be expressed as:

[0061] ω a =∫ T μ a dt=μ a T (4)

[0062] In the formula, μ a This is noise representing the rate of change of random relative acceleration.

[0063] The relative velocity noise can be obtained by integrating the relative acceleration noise over time, i.e.:

[0064]

[0065] Relative velocity noise integrated over time equals relative distance noise. Lateral and longitudinal distance noise ω q and ω p These are the components of the relative distance on the X and Y coordinate axes, respectively:

[0066]

[0067] Relative yaw rate noise It can be obtained by integrating the rate of change of random relative yaw acceleration over time, and can be expressed as:

[0068]

[0069] Relative yaw noise It can be expressed as the integral of the relative yaw rate, that is:

[0070]

[0071] In the formula, This is random relative yaw angle acceleration noise.

[0072] Finally, the process noise is expressed as:

[0073]

[0074] Furthermore, the measurement variables of the first vehicle state estimator and the second vehicle state estimator include the relative distance between the target vehicle and the vehicle itself, the rate of change of the relative distance, and the azimuth angle.

[0075] Millimeter-wave radar can accurately acquire information such as the relative distance, rate of change of relative distance, and azimuth angle of the target vehicle. Therefore, the measurement equation is defined as follows:

[0076] y k =h(x k )+δ k (11)

[0077] In the formula, Let ρ be the measurement vector. k The distance between the target vehicle and this vehicle in the current vehicle coordinate system; θ represents the rate of change of the relative distance between the target vehicle and the current vehicle. k Let be the angle between the target vehicle and the X-axis in the vehicle's own coordinate system; h(·) is the measurement equation, δ k The measurement noise is caused by distance measurement error and angle measurement error.

[0078] The measurement equation is used to calculate the predicted values ​​of the measured variables, and therefore should represent a function of the state variables. The relative distance between the target vehicle and the current vehicle satisfies the following relationship with the longitudinal and lateral distances in the state vector:

[0079]

[0080] The rate of change of the relative distance between the target vehicle and the current vehicle is obtained by differentiating equation (12) with respect to time, that is:

[0081]

[0082] The angle between the target vehicle and the X-axis in the vehicle's own coordinate system is obtained using the arctangent functions of the lateral and longitudinal distances, as follows:

[0083]

[0084] The final measurement equation is expressed as:

[0085] h(x k )=[h1(x k h2(x) k h3(x) k (15)

[0086] Furthermore, the warning method for a vehicle side collision warning system based on information fusion is as follows:

[0087] This invention proposes a method for predicting vehicle motion trajectory by using a grey model to predict GM(1,1) model for measurement variables and a square root capacitive Kalman filter algorithm for state correction.

[0088] Grey model prediction typically uses the GM(1,1) model, which can make advance predictions for processes with certain patterns of change. The calculation steps are as follows:

[0089] 1) Collect raw data and generate vector form

[0090] Y m =[y 0 (1),y 0 (2),y 0 (3),...,y 0 (m)] (16)

[0091] In the formula: m is the number of original data points taken.

[0092] 2) Accumulate the original data to generate a new sequence.

[0093]

[0094] 3) Generate data matrix

[0095]

[0096] 4) The development coefficient is calculated using the following formula. and

[0097]

[0098] 5) The corresponding whitening model is:

[0099]

[0100] The predicted sequence obtained after solving is as follows:

[0101]

[0102] 6) The predicted value obtained after restoration is:

[0103]

[0104] In the formula, n is the number of prediction steps in the grey model.

[0105] Grey model prediction uses limited historical information to accurately predict objects with regular changes, but its prediction accuracy is limited by the number of prediction steps and cannot achieve accurate prediction in multiple steps.

[0106] Therefore, this invention proposes a multi-step prediction method for vehicle motion trajectory using Kalman filtering that integrates grey model prediction, such as... Figure 2 As shown, after acquiring the relative motion state of the target vehicle, the first m-step state estimation is performed independently using the first square root capillary Kalman filter to achieve the relative motion state estimation of the vehicle. Then, starting from the m-th step, the vehicle trajectory prediction part begins. Using m sets of measurement vectors within time intervals k-m+1, k-m+2, ..., k, a grey model prediction is performed, with n prediction steps, obtaining the predicted measurement vector values ​​at times k+1, k+2, ..., k+n, which are then used as the actual measurement vectors for the second square root capillary Kalman filter. In the first step (k+1) trajectory prediction, the second square root capillary Kalman filter utilizes the estimated vehicle relative motion state values ​​from the k-step process. As the initial state, the grey model prediction value of the measurement vector As the measured vector, the relative motion state of the vehicle at step k+1 is estimated. Similarly, the second square root capacitive Kalman filter uses the vehicle relative motion state estimates from k+1 steps. As the initial state, the measurement vector is predicted by the grey model. Using the measured vectors, the relative motion state of the vehicle at step k+2 is estimated. Following this logic, the second square root capacitive Kalman filter ultimately estimates the relative motion state of the vehicles at k+n steps.

[0107] In the above vehicle trajectory prediction method: 1) The state transition equations and measurement equations of the two square root capacitive Kalman filters are the same, but because the trajectory prediction stage uses a grey model to predict multiple measurement vectors, while the vehicle state estimation stage only performs one state prediction, the parameters of the two square root capacitive Kalman filters are different; 2) The relative motion state of the vehicle at time k+n is predicted at time k. 3) If a collision is not expected, proceed to the next step of trajectory prediction. First, state estimation is performed at time k+1, then the vehicle trajectory at time k+n+1 is predicted. During trajectory prediction, each state vector obtained from state estimation is predicted one step using the state transition equation and then corrected by the measurement vector predicted by the grey model using the second square root occultary Kalman filter. Therefore, the method proposed in this invention solves the problem of poor adaptability of the motion model to uncertainties in driving conditions, improves prediction accuracy, and addresses the low accuracy of the grey model predictor during multi-step prediction.

[0108] Furthermore, after predicting the vehicle's trajectory, the vehicle collision warning module calculates the safe collision warning time, including the driver's reaction time, brake coordination time, deceleration increase time, and continuous braking time.

[0109] Vehicle braking process as Figure 5 As shown, the time includes driver reaction time t1, brake coordination time t2, deceleration increase time t3, and continuous braking time t4. Among these, t1 is influenced by factors such as driving skills and experience, and is generally 0.3–1.0 s; t2 is the time required to eliminate brake pedal clearance, and is generally 0.1–1.0 s; t3 is affected by the vehicle's brake performance. In this invention, t1 = t2 = t3 = 0.5 s is used.

[0110] The safe collision warning time refers to the time it takes for a vehicle to avoid a collision under maximum braking deceleration. This considers the road surface adhesion coefficient μ relative to the maximum deceleration -a. max After the influence of the effect, t4 can be expressed as:

[0111]

[0112] Then the safety collision warning time t min for:

[0113]

[0114] In the formula, v x Let g be the longitudinal relative speed, and g be the acceleration due to gravity.

[0115] Furthermore, the safe collision warning time of the lateral collision warning algorithm is corrected using the relative yaw angle. The safe collision warning time can be calculated given the road adhesion coefficient and relative speed. This invention assumes a collision occurs when the relative distance between the vehicles is less than 1 meter. If the trajectory prediction algorithm predicts a collision will occur within the safe collision warning time, a warning is issued. In the event of a lateral collision, the longitudinal relative speed is a component of the relative speed, and the safe collision warning time is corrected as follows:

[0116]

[0117] To ensure that the warning time is longer than the safe collision warning time, and considering algorithm error and latency, this invention defines the prediction time as t. min,k +0.5s, collision warning algorithm as follows Figure 3 As shown, at time k, after the signal acquisition system obtains the relevant information of the target vehicle relative to the vehicle itself, it uses a state estimation algorithm based on a constant turning rate and acceleration model using square root capacitance and Kalman filtering to obtain the motion state of the target vehicle relative to the vehicle itself. Based on information such as the current relative speed, relative yaw angle, and road adhesion coefficient, the safety collision warning time t is calculated. min,kThen, the trajectory prediction algorithm obtains the future value t. min,k The longitudinal and lateral distances of the target vehicle relative to this vehicle within +0.5s. If in the future t min,k If the square root of the sum of the squares of the longitudinal and lateral distances between the two vehicles within +0.5 seconds is less than 1 meter (S), a collision risk is identified, and a warning is issued. Conversely, if the distances are greater than 1 meter (S), the vehicle is deemed safe, and the process proceeds to the next moment, where t is calculated. min,k+1 Continue to predict subsequent t min,k+1 The vehicle trajectory is measured in +0.5s. This method takes into account the response time of the driver and brakes, as well as the road adhesion coefficient, ensuring that the driver or intelligent vehicle has sufficient braking time under different road conditions, thus improving vehicle safety.

[0118] The following example serves as an example to verify the effectiveness of the side collision warning method proposed in this invention. Figure 6 The collision scenarios shown include icy roads (μ=0.2), wet roads (μ=0.4), and dry roads (μ=0.6).

[0119] In a side-impact collision scenario, vehicle B merges into the main road from the ramp to the right of vehicle A. The initial longitudinal distance between vehicle A and vehicle B is 23.5m, and the lateral distance is 9m. Vehicle A's initial speed is 8m / s, and it is decelerating uniformly with an acceleration of -0.08m / s². Vehicle B's initial speed is 5.5m / s, and it is moving at a constant speed.

[0120] The safe collision warning time t calculated by equation (25) under different road surface adhesion coefficients. min like Figure 7 As shown. Because t min It is related to relative speed and relative yaw angle, therefore t for each road adhesion coefficient min It's not constant, but the smoother the road surface, the more tons are needed. min The larger. The actual and predicted relative distance between car B and car A, such as... Figure 8 As shown, the relative distance is defined as the arithmetic square root of the sum of the squares of the longitudinal and transverse distances. Figure 8 It is known that the two vehicles will collide at 9.70s, and when the road surface adhesion coefficient μ = 0.2, the collision warning system predicts a collision and issues a warning at 6.96s. Figure 7 It can be known that t at this time min = 2.68s. The early warning time under this condition is 9.70s - 6.96s = 2.74s, which is greater than t at this time. min Therefore, the warning requirements are met; when the road surface adhesion coefficient μ = 0.4, the collision warning system predicts a collision and issues a warning at 7.45s, at which time t min= 2.08s. The early warning time under this condition is 9.70s - 7.45s = 2.25s, which is greater than t at this time. min Therefore, the warning requirements are met; when the road surface adhesion coefficient μ = 0.6, the collision warning system predicts a collision and issues a warning at 7.56s, at which time t min = 1.88s. The early warning time under this condition is 9.70s - 7.56s = 2.14s, which is greater than t at this time. min Therefore, it meets the early warning requirements.

[0121] In summary, the collision warning method proposed in this invention can achieve early warning effects under different road surface adhesion coefficients. Compared to high-adhesion roads, the early warning time is longer on low-adhesion roads, ensuring that the driver or active obstacle avoidance control system can take timely control of the vehicle and improve driving safety.

[0122] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. For those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should be considered to fall within the scope of protection of the present invention.

Claims

1. A vehicle side collision warning system based on information fusion, characterized in that: The early warning system includes a signal acquisition module (20), a vehicle state estimation module (30), a vehicle trajectory prediction module (40), and a vehicle collision early warning module (50); The signal acquisition module (20) detects the relative position of the target vehicle (10) relative to the vehicle (60) and outputs it as a measurement variable to the first vehicle state estimator (31); The vehicle state estimation module (30) is the first vehicle state estimator (31), which is used to perform continuous m-step state estimation of the target vehicle relative to itself based on the collected relative measurement vector, as the raw data of the gray model predictor (41). The vehicle trajectory prediction module (40) includes a gray model predictor (41) and a second vehicle state estimator (42); The gray model predictor (41) performs n-step measurement prediction of GM(1,1) on the most recent m groups of measurement vectors, and inputs the obtained k+1 to k+n step measurement prediction values ​​as the actual measurement vectors of the second vehicle state estimator (42) in sequence. The second vehicle state estimator (42) estimates the relative motion trajectory of the vehicle and uses it as input to the vehicle collision warning module (50); The state variables of the first vehicle state estimator (31) and the second vehicle state estimator (42) include the lateral distance, longitudinal distance, relative speed, relative yaw angle, relative yaw rate and relative acceleration of the target vehicle (10) relative to the vehicle (60); The measurement variables of the first vehicle state estimator (31) and the second vehicle state estimator (42) include the relative distance of the target vehicle (10) to the vehicle (60), the rate of change of the relative distance, and the azimuth angle; The state transition equations of the first vehicle state estimator (31) and the second vehicle state estimator (42) are constant turning rate and acceleration models; The vehicle collision warning module (50) calculates the safe collision warning time t. min,k Inside, is the square root of the sum of the squares of the lateral and longitudinal distances of the target vehicle (10) relative to the vehicle (60) less than a critical value? If it is less than the critical value, it is determined that there is a risk of side collision, and a collision warning is issued to the vehicle (60). If the value is greater than the critical value, it is determined that there is no risk of side collision, and the next warning cycle begins; The vehicle collision warning module (50) calculates the safe collision warning time t. min,k This includes driver reaction time, brake coordination time, deceleration increase time, and continuous braking time, corrected for by relative yaw angle, and expressed as... Where k is the discrete time step, t min,k Let t1 be the safety collision warning time at the k-th warning step, t2 be the driver's reaction time and the brake coordination time, t3 be the acceleration increase time, and v be the acceleration increase time. x Let μ be the vehicle's longitudinal velocity, μ be the road surface adhesion coefficient, and g be the acceleration due to gravity. t4 is the relative yaw angle, and t4 is the continuous braking time, used to account for the influence of the road surface adhesion coefficient on the maximum deceleration. Before the relative yaw angle correction, t4 is given as v. x / (μg)-0.5t3.

2. The vehicle side collision warning system based on information fusion according to claim 1, characterized in that... The early warning methods used are as follows: After the signal acquisition module (20) acquires the relative motion state of the target vehicle (10), it first performs m-step state estimation independently through the first vehicle state estimator (31) to realize the relative motion state estimation of the vehicle. Then, starting from step m, we enter the vehicle trajectory prediction part; using m sets of measurement vectors in time k-m+1, k-m+2, ..., k, we perform gray model prediction, with n prediction steps, to obtain the predicted values ​​of the measurement vectors at time k+1, k+2, ..., k+n, and use them as the actual measurement vectors of the second vehicle state estimator (42). During the first step (k+1) trajectory prediction, the second vehicle state estimator (42) utilizes the vehicle relative motion state estimates from the k-step process. As the initial state, the grey model prediction value of the measurement vector As the measured vector, the relative motion state of the vehicle at step k+1 is estimated. The second vehicle state estimator (42) uses the vehicle relative motion state estimates from step k+1. As the initial state, the grey model prediction value of the measurement vector Using the measured vectors, the relative motion state of the vehicle at step k+2 is estimated. By analogy, the second vehicle state estimator (42) finally estimates the relative motion state of the vehicle in k+n steps.

3. The vehicle side collision warning system based on information fusion according to claim 1, characterized in that: The second vehicle state estimator (42) predicts the relative motion state of the vehicle at time k+n in the future. Used to determine whether a side collision will occur.

4. The vehicle side collision warning system based on information fusion according to claim 1, if it is determined at time k that a collision will not occur, then proceeding to the next step of trajectory prediction, characterized in that... The following methods were used to correct the measurement variables: First, perform state estimation at time k+1; Then predict the vehicle's trajectory at time k+n+1; During this process, the measurement variables of the grey model predictor (41) are corrected using the second vehicle state estimator (42).

Images