CN118144822A

CN118144822A - Rear-end collision prevention method for automatic driving

Info

Publication number: CN118144822A
Application number: CN202410436547.XA
Authority: CN
Inventors: 田永笑; 付睿; 刘阳华; 柳春; 王孟; 张泉; 彭艳
Original assignee: University of Shanghai for Science and Technology
Current assignee: University of Shanghai for Science and Technology
Filing date: 2024-04-11
Publication date: 2024-06-07

Abstract

The invention discloses an automatic driving-oriented rear-end collision prevention method, which comprises the steps of obtaining state information of a front vehicle relative to a self vehicle at the previous moment, inputting the state information into a motion model for input model interaction, and obtaining an interaction value; inputting the interaction value to the filter corresponding to each motion model for parallel filtering processing to obtain a filtering estimation result corresponding to each filter; updating the probability of each motion model according to the likelihood function of each filter and the semi-Markov state transition probability matrix to obtain a probability updating value of each motion model, wherein the state transition probability is a function of the residence time in the target state; predicting a state estimation value of the next moment of the front vehicle according to a filtering estimation result corresponding to each filter and a probability update value corresponding to each motion model; taking the predicted estimated value of the state of the front vehicle as the input interaction value of the next filtering process, and realizing real-time tracking of the front vehicle; and generating rear-end collision early warning information based on a real-time tracking result of the front vehicle.

Description

Rear-end collision prevention method for automatic driving

Technical Field

The invention relates to the technical field of maneuvering target tracking, in particular to an automatic driving-oriented rear-end collision prevention method.

Background

Today, the automatic driving technology is rapidly becoming a core competitive field of the automobile industry, especially in terms of improving road safety, and with the improvement of automation and intelligence level, the automatic driving technology plays an important role in preventing traffic accidents. Automobile accident studies by Mercees-Benz, germany, prove that: if the driver can recognize the danger and take measures 0.5 seconds earlier, 30% of head-on collision accidents and 50% of rear-end collision accidents can be avoided, and if the driver takes measures 1 second earlier, most accidents can be avoided. The automatic driving system aims at reducing human errors and improving driving safety through various sensors, calculation and control technologies, and a great deal of researches on rear-end collision early warning and avoiding systems at home and abroad for many years also show that the rear-end collision early warning system can reduce the braking response time and the rear-end collision quantity of a driver, and the rear-end collision reduction rate can reach 62%.

In the prior art, an interactive multi-model (INTERACTING MULTIPLE MODEL, IMM) target tracking algorithm can accurately estimate and predict a target motion state (such as position, speed, acceleration and the like), and can be used for maneuvering target tracking. Because the moment when the target is maneuvered is the moment when traffic accidents happen most easily, the tracking of the maneuvering target of the vehicle in the traffic field can reliably and accurately track and judge the target, effectively improve the driving safety of a driver and ensure the life and property safety. However, since a plurality of parallel Kalman filters are used in the interactive multimode, the improvement of tracking maneuver performance is replaced by larger computing resources, meanwhile, the accurate transition probability between models cannot be obtained in advance, and the use and tracking accuracy of the IMM algorithm are limited because the transition probability between models is determined in advance, the present statistical model of acceleration self-adaptive adjustment is combined with CV and CA models in the IMM algorithm in the patent application document with publication number CN102568004A, the Markov transition probability is adjusted by utilizing the system mode information hidden in the present measurement, the performance of the whole IMM algorithm is improved, the Markov transition probability is calculated on line in real time by utilizing the system mode information hidden in the present measurement, thus obtaining more accurate posterior estimation, and improving the precision of model fusion. However, this solution considers fewer models, only combining the "current" statistical model with the CV, CA model, and may still suffer from insufficient adaptability in the face of more complex or irregular maneuver targets. In the patent application document with publication number CN103853908a, it is proposed to construct likelihood ratio functions based on likelihood functions of constant-speed and uniform-acceleration motion models, and then calculate each variable in a markov transition matrix according to the likelihood ratio functions, and dynamically adjust the transition probability matrix depending on the calculation of the innovation, which may cause response delay in rapidly-changing maneuvering target tracking. The improvement of the transition probability matrix by the two is essentially in the category of a Markov chain, while in a standard Markov model, state transition is assumed to occur within a fixed time interval, which means that the opportunity of each transition is evenly distributed at each time point and has no dynamic adaptability.

Methods for preventing collision accidents also typically rely on radar and camera systems of the vehicle to monitor the surrounding environment, which systems are able to identify the presence of a vehicle in front and adjust the speed of travel based on inter-vehicle distance, relative speed and road conditions. However, these methods may not accurately predict and cope with emergency situations in complex or uncertain driving environments, especially in severe weather conditions or under limited sensors. For example, patent application publication CN109240310a proposes detecting the movement state of an obstacle and a vehicle by providing a plurality of sensors, analyzing these data to predict the movement track of the vehicle and the obstacle, comparing these tracks to evaluate the potential collision risk, and then automatically adjusting the movement of the vehicle to avoid the collision and to maintain a preset distance from the obstacle; however, the scheme excessively depends on the accuracy and the reliability of the sensor, and the fault or misreading of the sensor may cause poor obstacle avoidance effect; in extreme weather or complex traffic environments, detection and prediction of the obstacle may be inaccurate, affecting obstacle avoidance performance.

In addition, existing rear-end collision avoidance techniques focus primarily on immediate response and emergency handling, with less consideration to pre-prevent potential collisions through deep understanding of forward vehicle behavior. For example, current systems often do not take into account predictions of vehicle behavior patterns, such as acceleration, deceleration, lane changes, etc., which may limit the effectiveness of the rear-end collision avoidance system in complex traffic flows.

In view of the above limitations, it has become imperative to develop a more accurate and effective method of rear-end collision prevention. Such a method should be able to more accurately predict and understand the behavior of the preceding vehicle and take braking or other precautions if necessary to reduce or avoid the occurrence of rear-end collisions.

Disclosure of Invention

The technical problem to be solved by the invention is how to improve the adaptability and the accuracy of the automatic driving system in a changeable environment and improve the safety performance of the automatic driving vehicle in an actual road environment.

The invention solves the technical problems by the following technical means:

The invention provides an automatic driving-oriented rear-end collision prevention method, which comprises the following steps of:

acquiring state information of a front vehicle relative to a self vehicle at the previous moment, and inputting the state information into r motion models for input model interaction to obtain an interaction value;

Inputting the interaction value to the filter corresponding to each motion model for parallel filtering processing to obtain a filtering estimation result corresponding to each filter;

Updating the probability of each motion model according to the likelihood function of each filter and the semi-Markov state transition probability matrix to obtain a probability updating value of each motion model, wherein the state transition probability in the state transition probability matrix is a function of the stay time tau under the target state;

Predicting a state estimation value of the next moment of the front vehicle according to a filtering estimation result corresponding to each filter and a probability update value corresponding to each motion model;

Taking the predicted estimated value of the state of the front vehicle as the input interaction value of the next filtering process, and realizing real-time tracking of the front vehicle;

And generating rear-end collision early warning information based on a real-time tracking result of the front vehicle.

Further, the motion model includes a state equation and a measurement equation, wherein:

The state equation is:

X(k+1)＝FX(k)+GW(k)

The measurement equation is:

Z(k)＝H(k)X(k)+V(k)

Where X (k+1) is the target state vector at time k+1, r(k)、/>β(k)、Radial distance, relative velocity, relative acceleration, angle, angular velocity and angular acceleration at time k, respectively; f is a state transition matrix; g is a system noise matrix; w (k) is the system noise vector at time k; z (k) is observation data at time k,/>Z ₁(k)、z₂(k)、z₃ (k) is the distance, relative speed and azimuth angle of the front vehicle relative to the own vehicle measured at time k respectively; h (k) is the observation matrix; v (k) is the measurement noise.

Further, the step of obtaining the state information of the front vehicle relative to the own vehicle at the previous moment, and inputting the state information into r motion models for input model interaction to obtain an interaction value includes:

acquiring state information of a front vehicle relative to a self vehicle at the previous moment, wherein the state information comprises a distance, a relative speed and an angle;

processing the state information by using the motion model, and predicting the observation data at the current moment;

inputting the state information into r motion models respectively for input model interaction to obtain initial values of the r motion models after interaction, wherein the initial values comprise a filtering initial value and an estimated error covariance initial value;

And obtaining the interaction value according to the observed data and the initial value.

Further, the inputting the interaction value to the filter corresponding to each motion model to perform parallel filtering processing, to obtain a filtering estimation result corresponding to each filter, includes:

And taking the initial value output by each motion model as the input of the corresponding filter at the current moment and the observed data as the input of each filter, and then carrying out parallel filtering processing to obtain a filtering estimation result corresponding to each filter.

Further, the residence time τ in the target state is determined according to a residence time probability density matrix [ f _ij(τ_i ], τ _i is the residence time in state i, j representing the state.

Further, the updating the probability of each motion model according to the likelihood function and the state transition probability matrix of each filter to obtain a probability updating value of each motion model includes:

defining the probability that the dwell time equals τ at a moment k, the state i and known as Z (k), as a conditional probability mass function

Determining a conditional probability that the state becomes j at the time k under the condition that the state of the vehicle relative to the own vehicle before the time k-1 is i and the observation data is Z (k-1) based on the state transition probability p _ij (τ) and the conditional probability mass functionWherein p _ij(τ)＝P{M(k+τ)＝j|M(k)＝i,τ_i (k) =τ };

based on likelihood functions of the filters and the conditional probabilities And respectively updating the probability mu _i (k-1) of each motion model at the time k-1 to obtain a probability update value mu _j (k) of each motion model at the time k.

Further, the conditional probability mass functionGiven by the formula:

where a _i (k, 1) is the probability that the system will switch to state i at the next time step under observation Z (k-1) and in the non-state i condition; a _i (k, s) is the probability that the system remains in state i for consecutive s time steps under observation Z (k-s) and in non-state i; a _i (k, m) is the probability that the system remains in state i for successive m time steps under observation Z (k-m) and in non-state i; b _i (k, 1) represents the conditional probability that the system remains in state i for the next time step, given that the system is in state i at time k-1; b _i (k, s) represents the conditional probability that the system remains in state i for the next s time step given that the system is in state i at time k-s; b _i (k, m) is the conditional probability that the system remains in state i for the next m time steps given that the system is in state i at time k-m; mu _i (k-1) is the probability that the state is i at time k-1; mu _i (k-s) represents the probability that the state is i at time k-s; mu _i (k-m) represents the probability that the state is i at time k-m; The probability that the residence time is 1 time step number under the condition that the state at the moment k is i; /(I) The probability that the dwell time is equal to s given that the state at time k is i and Z (k) is known; /(I)Representing the probability that the system will remain in state i from the start of observation until time k and will continue to remain in state i until at least time k+1, under the condition of observation information Z (k) at time k.

Further, the conditional probabilityThe notation is:

Wherein, The probability that the dwell time equals τ at time k-1, the state i and known as Z (k-1), is a conditional probability mass function.

Further, the method further comprises:

judging whether the distance between the front vehicle and the own vehicle is smaller than the safe following distance or not based on the state information of the front vehicle and the own vehicle at the previous moment;

if yes, executing the step of inputting the state information into r motion models to perform input model interaction to obtain interaction values;

if not, the state information of the front vehicle relative to the own vehicle at the next moment is acquired again.

Further, the generating rear-end collision early warning information based on the real-time tracking result of the preceding vehicle includes:

based on the tracking result, judging whether the distance between the front vehicle and the own vehicle is smaller than the limit distance when the threat is determined during the transverse collision and the longitudinal collision;

if yes, triggering active braking;

If not, triggering rear-end collision early warning when the distance between the front vehicle and the own vehicle is smaller than the guard distance.

The invention has the advantages that:

(1) The invention considers that in actual road conditions, the vehicle behavior (such as acceleration, deceleration, lane change, etc.) does not always follow a fixed time interval, so that an interactive multi-model algorithm with a motion state following a semi-markov chain can be adopted, and the key feature of the semi-markov chain is that the concept of 'stay time' (sojoum time) is introduced, namely the stay time of the system in the current state, which can be determined according to a specific probability distribution, can effectively process the state transition of the non-fixed time interval, capture more complex system dynamics, such as the duration of the state affects the following state transition, so that the model is more consistent with the running characteristics of real conditions and actual systems, and can adjust the behavior of an automatic driving vehicle in real time under the changeable road conditions according to the change of real-time road and traffic conditions, thereby more accurately predicting and responding to the potential rear-end collision risk.

(2) By analyzing the longitudinal and transverse movement states among vehicles, the potential rear-end collision risk is accurately judged, and the early warning and emergency response capabilities of the system are improved.

Additional aspects and advantages of the invention will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the invention.

Drawings

Fig. 1 is a schematic flow chart of an automatic driving-oriented rear-end collision prevention method according to an embodiment of the present invention;

FIG. 2 is a block flow diagram of an automatic driving-oriented rear-end collision prevention method according to an embodiment of the present invention;

fig. 3 is a block diagram of a system implementation corresponding to an automatic driving-oriented rear-end collision prevention method according to an embodiment of the present invention.

Detailed Description

For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions in the embodiments of the present invention will be clearly and completely described in the following in conjunction with the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

As shown in fig. 1, an embodiment of the present invention provides an automatic driving-oriented rear-end collision prevention method, which includes the following steps:

s10, acquiring state information of a front vehicle relative to a self vehicle at the previous moment, and inputting the state information into r motion models for input model interaction to obtain an interaction value;

Specifically, the present embodiment may employ a millimeter wave pulse doppler radar sensor, GPS, a vision camera, and the like to collect state information of the vehicle; the millimeter wave pulse Doppler radar sensor not only can measure the distance relative speed of a front target, but also has angle measurement capability, and a large number of experimental researches at home and abroad show that the azimuth information of the target is essential for removing false alarms; while other sensors such as GPS and vision cameras may assist in target detection and localization.

S20, inputting the interaction value into the filter corresponding to each motion model to perform parallel filtering processing, and obtaining a filtering estimation result corresponding to each filter;

It should be noted that, the interactive multimode includes a plurality of motion models, in this embodiment, the number of filters is the same as that of the motion models and corresponds to one another, each motion model uses an independent filter, and the filtering process is that the plurality of filters perform parallel computation according to different motion models.

S30, updating the probability of each motion model according to the likelihood function of each filter and the semi-Markov state transition probability matrix to obtain a probability update value of each motion model, wherein the state transition probability in the state transition probability matrix is a function of the stay time tau under the target state;

It should be noted that the half markov state transition probability matrix in this embodiment is different from the standard markov state transition probability matrix, and the state transition in the standard markov model is assumed to occur within a fixed time interval, which means that the opportunity of each transition is evenly distributed at each time point. The key feature of the semi-markov chain designed in this embodiment is that it introduces the concept of "dwell time", i.e. the time the system stays in the current state, which can be determined according to a specific probability distribution, which allows the algorithm to handle state transitions at non-fixed time intervals, which can describe the transitions of the object between different states more accurately by considering the duration of the states and finer state transition modeling, thereby improving tracking accuracy.

S40, predicting a state estimation value of the next moment of the front vehicle according to a filtering estimation result corresponding to each filter and a probability update value corresponding to each motion model;

S50, taking the predicted estimated value of the state of the front vehicle as an input interaction value of the next filtering process, and realizing real-time tracking of the front vehicle;

S60, generating rear-end collision early warning information based on a real-time tracking result of the front vehicle.

The SMC-IMM algorithm provided by the embodiment combines an IMM algorithm and a Semi-Markov Chain (Semi-Markov Chain), can process state conversion at a non-fixed time interval, and can quickly adjust the behavior of an automatic driving vehicle according to the change of real-time road and traffic conditions in combination with the consideration of dynamic adaptability, so that the potential rear-end collision risk can be predicted and responded more accurately.

It should be noted that, as a dynamic system, the vehicle may exhibit different motion patterns (such as uniform motion, acceleration motion, or turning motion) during its operation, and in order to accurately estimate the states of these systems, the IMM algorithm uses a plurality of motion models to cover these different motion patterns. The state equations of the motion model equations describe the evolution of the system states (such as position, velocity, and acceleration) over time, among others. Different state equation models (e.g., uniform linear motion, uniform acceleration motion, etc.) mathematically describe how these state variables change over time; the metrology equation of the motion model equation describes how observables (e.g., position and velocity data acquired by radar or GPS) are derived from the system state. These equations take into account errors and noise in the observation process, enabling algorithms to combine actual observation data with model predictions to improve estimation accuracy.

Specifically, the process for establishing the motion model comprises the following steps:

The dynamic model describes and estimates motion parameters related to the prediction process in the form of differential equations, and the motion of an object in 3D space can be described as a combined result of rotation and translation. Since the longitudinal slope of a highway is small, it can be approximated as horizontal over a short distance, it is assumed herein that the road plane is horizontal, i.e. the road has no undulations in the vertical direction. On the assumption that, depending on the dynamics of the vehicle motion, two directions of freedom of the target motion are to be considered: 1) The target rotates around the normal line of the ground, so that the moving direction of the target is not parallel to the road direction; 2) Translation along the road direction.

1) Establishing a state equation

For tracking moving objects, most important is the distance r between the object and the radar and the rate of change of the distance, i.e. the relative movement speedThe relative motion speed is the derivative of the relative distance with respect to time. Lateral movement angular velocity of target front truck relative to radar/>And the angle beta determines the position of the target on the road at the next moment, which is important for the running of the vehicle. The millimeter wave pulse Doppler radar used in the embodiment of the invention measures the distance, the relative speed and the angle of the front vehicle, and the signals to be estimated are the distance r, the relative speed/> of the target at the next momentAngle beta.

Let the distance, speed and angle of the target at time k (when the radar scans a certain time) be r (k),Beta (k), the distance, speed and azimuth at time (k+1) (at the next scan of the radar) are r (k+1),/>, respectivelyAnd β (k+1). If the time T of two adjacent scans by the radar is small enough, then it can be approximately considered that:

Wherein: And/> The radial acceleration and the azimuth change speed of the target at the time k are respectively.

It is assumed that radial acceleration and azimuth change acceleration of the target are randomly changed due to random factors such as sudden gusts, road conditions, irregular changes in road adhesion coefficient, maneuvering operation of the driver, and the like:

Wherein: u ₁ (k) is zero in mean and variance U ₂ (k) is the increment of the radial acceleration and azimuth change speed from time k to time (k+1), which is an uncorrelated random white noise sequence with u ₁ (k); /(I)Representing the relative acceleration at time k,/>The angular acceleration at time k is indicated.

The state equation of the target motion at this time can be described as follows:

X(k+1)＝FX(k)+GW(k)

Wherein:

W(k)＝[w₁(k) w₂(k)]

Wherein: f is a state transition matrix, G is a system noise matrix, T is a sampling interval, a variable X (k) is a target state vector, r (k), β(k)、/>The radial distance, relative velocity, relative acceleration, angle, angular velocity and angular acceleration at time k, respectively. W (k) is a system noise vector, W ₁(k)、w₂ (k) is a random white noise sequence with mean value of zero and uncorrelated, and its variance is/>Σ _a is the variance of w ₁ (k), σ _β is the variance of w ₂ (k), and the specific values need to be determined according to the system characteristics and design requirements in practical applications, and are usually set by experiment or based on the system performance requirements.

The covariance matrix of W (k) is:

2) Establishing a measurement equation

By adopting the vehicle millimeter wave radar, the measurement equation of the distance, the relative speed and the azimuth angle of the target can be written as follows:

Z(k)＝H(k)X(k)+V(k)

Wherein:

Wherein: z ₁(k)、z₂(k)、z₃ (k) represents the distance, relative velocity and azimuth of the target as measured by the radar, respectively; h (k) is the observation matrix; v (k) is measurement noise; wherein v ₁(k)、v₂(k)、v₃ (k) is three mutually uncorrelated random white noise sequences, the equations of which are respectively And/>

Wherein, the covariance matrix of V (k) is:

Here, the Is the variance of the distance measurement error,/>Is the variance of the error of the relative velocity measurements,/>Is the variance of the azimuth measurement error.

Further, an interactive multi-model (SMC-IMM) algorithm is designed that follows a semi-Markov chain, the process being as follows:

One of the features of a semi-markov chain: there is a fixed transition probability matrix [ p _ij ] and a dwell time probability density matrix [ f _ij(τ_i) ] that are functions of the current state i and the target state j, and in a semi-markov process the transition time of the system can follow different types of probability distributions, the choice of which depends on the nature of the target process and the actual observation data. In a semi-Markov chain, the target state of a jump is first selected based on [ p _ij ], and then the time after the jump occurs (i.e., the dwell time) is selected based on the dwell time probability density matrix [ f _ij(τ_i ]. And two features: it is the transition probability p _ij that is a function of the dwell time τ, whereas in conventional IMM p _ij is a fixed constant set according to a priori information.

To obtain p _ij (τ) in the semi-markov chain, where p _ij(τ)＝P{M(k+τ)＝j|M(k)＝i,τ_i (k) =τ; the present embodiment first defines the probability that the dwell time equals τ at time k, the state i and known as Z (k), as a conditional probability mass function

In the method, in the process of the invention,This notation "defined as"; τ _i (k) represents the duration or dwell time of state i at time k; m (k) represents the system state at time k; m (k-1) represents the system state at time k-1; m (k- τ+1) represents the system state for τ -1 time steps forward from the current time k; z (k) represents observation data at time k; p { τ _i (k) =τ|m (k) =i, Z (k) } is the probability that the system stays in state i for a time equal to τ given that the current time k is in state i and the measured value Z (k); p { τ _i (k) =τ|m (k) =i, Z (k-1) } is the probability that the system stays in state i for a time equal to τ given that the current time k is in state i and the measured value Z (k-1); p { M (k-1) =i, …, M (k- τ+1) =i, M (k- τ) +.i|m (k) =i, Z (k-1) } represents the probability that the system was in state i all the time from k- τ+1 to k-1, but not in state i at k- τ, at a known time k in state i.

Wherein the conditional probability mass function of the dwell time τ of time k in state i is given by:

In the above formula, a _i (k, 1) is the probability that the system will switch to state i at the next time step under the condition that the observed data Z (k-1) is in non-state i; a _i (k, s) is the probability that the system remains in state i for consecutive s time steps under observation Z (k-s) and in non-state i; a _i (k, m) is the probability that the system remains in state i for successive m time steps under observation Z (k-m) and in non-state i; b _i (k, 1) represents the conditional probability that the system remains in state i for the next time step, given that the system is in state i at time k-1; b _i (k, s) represents the conditional probability that the system remains in state i for the next s time step given that the system is in state i at time k-s; b _i (k, m) is the conditional probability that the system remains in state i for the next m time steps given that the system is in state i at time k-m; mu _i (k-1) is the probability that the state is i at time k-1; mu _i (k-s) represents the probability that the state is i at time k-s; mu _i (k-m) represents the probability that the state is i at time k-m; The probability that the residence time is 1 time step number under the condition that the state at the moment k is i; /(I) The probability that the dwell time is equal to s given that the state at time k is i and Z (k) is known; /(I)Representing the probability that the system will remain in state i from the start of observation until time k and will continue to remain in state i until at least time k+1, under the condition of observation information Z (k) at time k.

[ A _i ] and [ b _i ] are defined as the probability of staying in the same state for s time steps, conditioned on the information at k-s, given by:

wherein: z (k-s) represents the observed data at time k-s; p _ii (j) represents the probability of remaining in state i; the probability that the dwell time is equal to n under the condition that the state is i at the moment of k-s; n represents different time points.

On the condition that the available information Z (k-s) at time k-s, the probability of being in the same state i in the next s time steps is:

Where P { M (k) =i, …, M (k-s+1) =i|z (k-s) } is the probability that the system is in state i from time k-s+1 to time k under the condition of the measured value Z (k-s) at time k-s; τ _j (k-s) represents the duration of state j before time k-s; p _ji (n) is the transition probability of transitioning from state j to state i at time n; l represents the time step in the successive product operation.

Defining the conditional probability of the state at time k-1 to become j under the conditions that the state at time k-1 is i and the measured value is Z (k-1)

In one embodiment, the step S10: acquiring state information of a front vehicle relative to a self vehicle at the previous moment, inputting the state information into r motion models for input model interaction to obtain an interaction value, and comprising the following steps:

s11, acquiring state information of a front vehicle relative to a self vehicle at the previous moment, wherein the state information comprises a distance, a relative speed and an angle;

S12, processing the state information by using the motion model, and estimating observation data at the current moment;

It should be noted that, in this embodiment, the observation data can be presumed by using the measurement equation.

S13, respectively inputting the state information into r motion models to perform input model interaction, and obtaining initial values of the r motion models after interaction, wherein the initial values comprise a filtering initial value and an estimated error covariance initial value;

S14, obtaining the interaction value according to the observed data and the initial value.

Specifically, the present embodiment may calculate, according to the filtered values and the model probabilities at the time of two models (k-1), the filtered initial values after the interactive mixing, including the filtered initial values of model 1: filtering an estimateAnd estimating covariance μ ¹ (k-1); filter initial value of model 2: filtering estimate/>And estimation error covariance/>

Let the probability of model 1 at time (k-1) of the system be mu ¹ (k-1), the filtered valueThe estimated error covariance is P ² (k-1). Model 2 has a probability of μ ² (k-1), a filter value of X ² (k-1), and a system estimation error covariance of P ² (k-1). Then further spreading the filter initial values to r models, wherein the filter initial values of the r motion models after interaction are as follows:

Wherein:

Wherein: To estimate the state of the model at time k-1, based on the state estimate of the i-th model, For the mixed initial state estimation value calculated for the jth model during the interaction of time k-1 to k-1, P ^0j (k-1|k-1) is the mixed initial estimation error covariance calculated for the jth model, mu _ij (k-1|k-1) is the mixed probability from model i to model j at time k-1, P ⁱ (k-1|k-1) is the estimated error covariance based on the ith model at time k-1, mu _ij (k-1|k-1) is the same as above, mu _i (k-1) is the model probability of the ith model at time k-1,/>In order to obtain a conditional probability that the state becomes j at time k under the condition that the state is i at time k-1 and the measured value is Z (k-1), p _ij (τ) is a state transition probability,The probability that the dwell time equals τ at time k-1, the state i and known as Z (k-1), is a conditional probability mass function.

It should be noted that the number of the substrates,Is a direct estimate of the state of the system at time k-1 based on the filtered value of model 1, i.e. given model 1. /(I)Is the hybrid initial state estimate calculated for model j during the interaction between time k-1 and k-1. This value is obtained by mixing the estimated values of the different models at time k-1 according to a certain probability weight for the next filtering process under model j. /(I)

Further, the method comprises the steps of,Is an estimate of the transition probability of model i to model j,/>Is a normalization constant.

In one embodiment, the step S20: inputting the interaction value to the filter corresponding to each motion model for parallel filtering processing to obtain a filtering estimation result corresponding to each filter, wherein the method specifically comprises the following steps:

For example, a filter M _j (k) corresponding to the model takes X ^0j(k-1|k-1),P^0j (k-1|k-1) and Z (k) as inputs to the filter corresponding to the motion model for Kalman filtering, wherein:

Kalman prediction equation:

Prediction error covariance matrix:

Kalman gain:

K_j(k)＝P^j(k|k-1)H^T[HP^j(k|k-1)H^T+R]^-1

The filter equation is:

filtering error variance matrix:

P^j(k|k)＝[I-K_j(k)H]P^j(k|k-1)

Wherein, Under the representation model j, on the basis of the information of the moment k-1, a state prediction value of the moment k is obtained; a ^j represents the state transition matrix under model j; g _j denotes the control input matrix under model j; q _j represents the system noise covariance matrix under model j; p ^j (k|k-1) represents a state prediction error covariance matrix of the time k on the basis of information of the time k-1 under the model j; k _j (K) represents the kalman gain at time K under model j; r represents an observation noise covariance matrix; under the representation model j, estimating the optimal state of the time k on the basis of the information of the time k; p ^j (k|k) represents an error covariance matrix of the state estimation based on the information of the moment k under the model j; i represents an identity matrix.

In one embodiment, the step S30: updating the probability of each motion model according to the likelihood function and the semi-Markov state transition probability matrix of each filter, wherein the probability is specifically as follows:

wherein the likelihood function Λ _j (k):

Wherein: mu _j (k) is the probability of the jth motion model at time k; z (k) is the observed data at time k, i.e., the value measured by the sensor; Is a conditional probability density function of the observed data Z (k) given the state estimation and covariance of the jth model and the last moment, the probability indicating the probability that the observed data Z (k) appears in the jth model; p [ Z (k) |m (k) =j, Z (k-1) ] is the conditional probability of the observation data Z (k) under the j-th model under the condition of the observation data Z (k-1).

In one embodiment, in the step S40: according to the filtering estimation results corresponding to the filters and the probability update values corresponding to the motion models, predicting the state estimation value of the next moment of the front vehicle as follows:

Wherein, The state estimation value after all the observation information up to the time k is given at the time k, and the state estimation value is the best estimation of the current state of the front vehicle; p (k|k) represents an estimation error covariance matrix; mu _j (k) is the probability of the jth model at time k, representing the likelihood that the model is applicable to the current state; p ^j (k|k) is the estimated error covariance matrix corresponding to time k under the jth model; /(I)Is an estimate of the state of the preceding vehicle at time k under the j-th model.

In an embodiment, as shown in fig. 2, the method further comprises the steps of:

In one embodiment, the step S60: the method for generating rear-end collision early warning information based on the real-time tracking result of the preceding vehicle comprises the following steps:

if yes, triggering active braking;

Specifically, the tracking result of the front vehicle includes the inter-vehicle distance, the speed, the acceleration and the transverse angle, the embodiment removes the false alarm according to the transverse angular speed and the distance, judges whether rear-end collision is possible according to the longitudinal movement state, and the excessively high false alarm rate can not only reduce the workload of the driver, but also cause the mental height of the driver to be stressed, thereby achieving the opposite effect.

Further, consider the following possible scenario: the radar detects the road side guard rail and the like, and false alarm may be generated, so that alarm suppression is required. The distance between the two vehicles may be smaller than the safe following distance, and in fact, because the front vehicle has a certain transverse speed relative to the own vehicle, when the own vehicle approaches the front vehicle, the front vehicle may deviate to a safe transverse position, and the alarm system will generate false alarm, and the alarm should be restrained

Further, according to the detected azimuth information of the front vehicle and the own vehicle, estimating the transverse angular speed of the front vehicle relative to the own vehicle in real time according to the millimeter wave measurement value, calculating the transverse distance of the front vehicle relative to the own vehicle at the next moment, and confirming whether a potential threat is formed to the own vehicle or not, namely, effectively identifying a target object; and then, judging whether rear-end collision is possible or not based on the longitudinal collision judgment of the tracking result. Specifically, v ₁ represents the speed of the vehicle, and v _rel represents the relative speed. If v _rel>v₁, the target is considered to be located on the opposite lane, and longitudinal collision judgment is not needed.

The millimeter wave radar can measure the distance r and the angle beta between two vehicles, and the distance r and the change rate of the distance between the front vehicle and the radar, namely the relative movement speed r, the change rate of r, namelyFor relative acceleration,/>Is accelerating,/>Negative is deceleration. Lateral movement angular velocity of front vehicle relative to radar/>And the angle beta determines the position of the front vehicle on the road at the next moment, and whether the road is changed or not can be predicted.

Further, for the case where the target object is located on the adjacent lane, the lateral distance value b=sinβ between the preceding vehicle and the own vehicle can be calculated by a simple geometric operation relationship using the angle β of the preceding vehicle with respect to the own vehicle axis (i.e., the azimuth angle of the preceding vehicle measured by the radar) and the distance r between the preceding vehicle and the own vehicle obtained from the information transmitted from the radar to the processing unit. After the lateral distance b is obtained, it is compared with the vehicle width c ₁ and the front vehicle width c ₂. If b > (c ₁+c₂)/2, it can be concluded that the preceding vehicle is not on the driving route of the own vehicle and does not pose a threat to the operation of the own vehicle; if b.ltoreq.c ₁+c₂/2, then the preceding vehicle has some up to all on the path of travel of the own vehicle, which is a potential threat to the own vehicle.

Further, since the radar measurement data contains azimuth information, it is necessary to extract the information of the lateral relative velocity between the vehicles from the azimuth information. If the direct differentiation is adopted on the azimuth angle signal to obtain the transverse relative speed value, the method is not feasible because of the following two points:

One of the reasons is that the target azimuth signal transmitted by the radar contains random errors, the influence of the errors on the transverse relative speed is large, and the transverse relative speed value obtained by actual calculation is difficult to apply.

The second reason is that the real-time performance of the system is poor due to the limitation of the algorithm (differentiation of data, and certain time interval is needed between data). Further, smooth estimation of the system state vector can be obtained by adopting SMC-IMM, and the method is applied to radar data processing of an automobile rear-end collision early warning system so as to make up for the two defects.

Specifically, the safety distance to be maintained is compared with the measured inter-vehicle distance. If the measured workshop distance is greater than the reminding alarm distance, entering the next working cycle; if the measured workshop distance is smaller than the warning distance, giving an alarm once to remind a driver to loosen the accelerator and brake; and when the actual measurement workshop distance is smaller than the limit distance, executing active braking so as to avoid rear-end collision accidents.

Preferably, the limit distance isWherein a _m is the relative acceleration of the maximum braking force which can be provided after thermal decay, and the braking margin d ₀ is more than or equal to 2.0m.

Further, as shown in fig. 3, the embodiment of the invention further provides a system structure for implementing the method, which comprises a detection unit, an information processing unit and an execution unit, wherein the detection unit is used for measuring and acquiring the state of a front vehicle and the motion state of a self vehicle through a plurality of sensors; the information processing unit analyzes and establishes a state equation and a measurement equation of a motion model of the system, and the SMC-IMM algorithm carries out target detection and real-time tracking on a front vehicle; the execution unit is used for giving early warning or emergency braking. The system combines sensor data and algorithm processing capability, and can adjust the behavior of the automatic driving vehicle in real time under changeable road conditions.

In the description of the present specification, a description referring to terms "one embodiment," "some embodiments," "examples," "specific examples," or "some examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiments or examples. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

Furthermore, the terms "first," "second," and the like, are used for descriptive purposes only and are not to be construed as indicating or implying a relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include at least one such feature. In the description of the present invention, the meaning of "plurality" means at least two, for example, two, three, etc., unless specifically defined otherwise.

While embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and not to be construed as limiting the invention, and that variations, modifications, alternatives and variations may be made to the above embodiments by one of ordinary skill in the art within the scope of the invention.

Claims

1. An automatic driving-oriented rear-end collision prevention method, comprising:

2. The autopilot-oriented rear-end collision avoidance method of claim 1 wherein the motion model comprises a state equation and a measurement equation, wherein:

The state equation is:

X(k+1)＝FX(k)+GW(k)

The measurement equation is:

Z(k)＝H(k)X(k)+V(k)

3. The method for preventing rear-end collision for automatic driving according to claim 1, wherein the steps of obtaining the state information of the preceding vehicle relative to the own vehicle at the previous moment, and inputting the state information into r motion models for input model interaction, and obtaining the interaction value comprise:

4. The method for preventing rear-end collision for automatic driving according to claim 1, wherein the step of inputting the interaction value to the filter corresponding to each motion model to perform parallel filtering processing to obtain a filtering estimation result corresponding to each filter comprises the steps of:

5. The method for rear-end collision avoidance as claimed in claim 1, characterized in that the residence time τ in the target state is determined from a residence time probability density matrix [ f _ij(τ_i ], τ _i is the residence time in state i, j representing the state.

6. The method for preventing rear-end collision according to claim 1, wherein updating the probability of each motion model according to the likelihood function and the state transition probability matrix of each filter to obtain the probability update value of each motion model comprises:

Determining a conditional probability that the state becomes j at time k under the condition that the state is i at time k-1 and the observed data is Z (k-1) based on the state transition probability p _ij (τ) and the conditional probability mass function

Based on likelihood functions of the filters and the conditional probabilitiesAnd respectively updating the probability mu _i (k-1) of each motion model at the time k-1 to obtain a probability update value mu _j (k) of each motion model at the time k.

7. The automatic driving oriented rear-end collision prevention method as claimed in claim 1, characterized in that said conditional probability mass functionGiven by the formula:

Wherein [ a _i ] and [ b _i ] are defined as probabilities of staying in the same state for s time steps, on the condition that information at k-s is present; a _i (k, 1) is the probability that the system will switch to state i at the next time step under observation Z (k-1) and in the non-state i condition; a _i (k, s) is the probability that the system remains in state i for consecutive s time steps under observation Z (k-s) and in non-state i; a _i (k, m) is the probability that the system remains in state i for successive m time steps under observation Z (k-m) and in non-state i; b _i (k, 1) represents the conditional probability that the system remains in state i for the next time step, given that the system is in state i at time k-1; b _i (k, s) represents the conditional probability that the system remains in state i for the next s time step given that the system is in state i at time k-s; b _i (k, m) is the conditional probability that the system remains in state i for the next m time steps given that the system is in state i at time k-m; mu _i (k-1) is the probability that the state is i at time k-1; mu _i (k-s) represents the probability that the state is i at time k-s; mu _i (k-m) represents the probability that the state is i at time k-m; The probability that the residence time is 1 time step number under the condition that the state at the moment k is i; /(I) The probability that the dwell time is equal to s given that the state at time k is i and Z (k) is known; /(I)Representing the probability that the system will remain in state i from the start of observation until time k and will continue to remain in state i until at least time k+1, under the condition of observation information Z (k) at time k.

8. The method for preventing rear-end collision according to claim 6, wherein said conditional probabilityThe notation is:

9. The autopilot-oriented rear-end collision avoidance method of claim 1 wherein said method further comprises:

10. The method for preventing rear-end collision for automatic driving according to claim 1, wherein the generating rear-end collision early warning information based on the real-time tracking result of the preceding vehicle comprises:

if yes, triggering active braking;