Vehicle driving behavior recognition method and device
Through the Hidden Markov HMM model, the lane change behavior of vehicles is predicted, and the lane keeping, left lane change and right lane change prediction models are used to solve the problem that autonomous vehicles cannot identify lane change intentions in advance, and improve the recognition accuracy and safety .
Patent Information
- Application Number
- CN202111247701.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-10-26
- Publication Date
- 2024-12-24
- Estimated Expiration
- 2041-10-26
AI Technical Summary
Self-driving cars cannot identify the drivers of surrounding vehicles in advance, resulting in automatic operation errors and increasing the risk of traffic accidents.
The lane keeping prediction model, left lane change prediction model and right lane change prediction model are adopted to train and predict using the Hidden Markov HMM model by obtaining the vehicle's observed motion parameters (lateral displacement, lateral velocity and yaw angular velocity). Output lane keeping probability, left lane change probability and right lane change probability to judge vehicle behavior trend.
Accurate and early identification of the intention to change lanes by surrounding vehicles is achieved, reducing autonomous driving errors and avoiding traffic accidents.
Smart Images
Figure CN114022855B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of autonomous driving technology, and in particular to a vehicle driving behavior recognition method and device. Background Art
[0002] Autonomous vehicles (Self-driving automobile), also known as driverless cars, computer-driven cars, or wheeled mobile robots, are intelligent cars that are unmanned by computer systems. Self-driving cars rely on artificial intelligence, visual computing, radar, monitoring devices, and global positioning systems to work together, allowing computers to automatically and safely operate motor vehicles without any human intervention. Unmanned driving technology integrates many high technologies such as artificial intelligence, computer vision, integrated navigation, information fusion, automatic control, and architecture, and is the product of the high development of computer science and automation technology.
[0003] Perception and cognition are one of the most critical links in autonomous driving and are the prerequisite for correct planning and decision-making. For example, lane changing and sudden braking of surrounding vehicles. Especially lane changing behavior, its changing trend is slow and difficult to identify and judge.
[0004] However, in the highway environment, if the lane-changing intention of the drivers of surrounding vehicles cannot be identified in advance, it will lead to automatic operation errors and more likely cause serious traffic accidents. The current technical methods for judging the lane-changing behavior of surrounding vehicles are not ideal. Either the recognition is inaccurate, or the amount of data required is large, the processing is complicated, the calculation is too large, and a high-power processor is required, which increases the cost of vehicle manufacturing. Summary of the invention
[0005] In view of the defects existing in the prior art, the purpose of the present invention is to provide a vehicle driving behavior recognition method, which can solve the problem in the prior art that automatic driving cannot identify the lane changing intention behavior of drivers of surrounding vehicles in advance, which may lead to automatic operation errors and more easily cause serious traffic accidents.
[0006] In order to achieve the above purpose, the technical solution adopted by the present invention is:
[0007] In one aspect, the present invention provides a vehicle driving behavior recognition method, comprising:
[0008] Obtaining observed motion parameters of the target vehicle;
[0009] Inputting the observed motion parameters into a lane keeping prediction model, a left lane change prediction model, and a right lane change prediction model respectively;
[0010] The vehicle behavior trend is determined according to the lane keeping probability, left lane changing probability and right lane changing probability output by the lane keeping prediction model, the left lane changing prediction model and the right lane changing prediction model.
[0011] In some optional schemes, before obtaining the observed motion parameters of the target vehicle, it also includes training a lane keeping prediction model, a left lane change prediction model and a right lane change prediction model.
[0012] In some optional solutions, the methods of training the lane keeping prediction model, the left lane change prediction model, and the right lane change prediction model all include:
[0013] Establishing an initial lane-changing behavior prediction model for each lane-changing behavior;
[0014] Collect the observed motion parameters corresponding to each lane-changing behavior of the vehicle in actual traffic scenarios;
[0015] Based on the Baum-Welch algorithm, the observed motion parameters corresponding to each are sequentially introduced into the initial lane changing behavior prediction model corresponding to the lane changing behavior for training, and the trained lane keeping prediction model, left lane change prediction model and right lane change prediction model are obtained.
[0016] In some optional solutions, the Baum-Welch algorithm is based on sequentially introducing the observed motion parameters corresponding to each of the initial lane changing behavior prediction models corresponding to the lane changing behavior for training, so as to obtain the trained lane keeping prediction model, the left lane change prediction model and the right lane change prediction model, including:
[0017] Substituting the observed motion parameters corresponding to each into the initial lane-changing behavior prediction model corresponding to the lane-changing behavior in turn, and obtaining the joint distribution function of each iterative lane-changing behavior prediction model;
[0018] The expectation of each joint distribution function is maximized to obtain an updated iterative lane changing behavior prediction model until the state transition probability in the iterative lane changing behavior prediction model converges. The obtained iterative lane changing behavior prediction models are used as the trained lane keeping prediction model, left lane change prediction model and right lane change prediction model.
[0019] In some optional schemes, the training data of the lane keeping prediction model, the left lane change prediction model and the right lane change prediction model all include lane change status, observed motion parameters, initial state distribution probability, state transition probability, observed motion parameter mean and observed motion parameter variance.
[0020] In some optional schemes, the observed motion parameters include the lateral displacement of the vehicle relative to the center line of the lane within a set time period, the lateral speed and the vehicle yaw rate.
[0021] In some optional schemes, the lane changing status includes lane changing start, lane changing in progress and lane changing end.
[0022] In some optional solutions, the state transition probability is the probability of a certain lane changing state at a certain moment transitioning to any lane changing state at the next moment.
[0023] In some optional schemes, the lane changing behavior includes left lane changing, lane keeping and right lane changing.
[0024] On the other hand, the present invention also provides a vehicle driving behavior recognition device, comprising:
[0025] A parameter acquisition module, which is used to obtain the observed motion parameters of the target vehicle;
[0026] A calculation and analysis module, which is used to input the observed motion parameters into a lane keeping prediction model, a left lane change prediction model, and a right lane change prediction model respectively;
[0027] A judgment output module is used to judge the vehicle behavior trend according to the lane keeping probability, left lane changing probability and right lane changing probability output by the lane keeping prediction model, the left lane changing prediction model and the right lane changing prediction model.
[0028] Compared with the prior art, the advantages of the present invention are: when using the optimized left lane change prediction model, lane keeping prediction model and right lane change prediction model, firstly, the observed motion parameters of the target vehicle are obtained, and the observed motion parameters are selected from the lateral displacement, lateral velocity and vehicle yaw angular velocity of the target vehicle relative to the lane centerline, and the observed motion parameters are respectively input into the lane keeping prediction model, the left lane change prediction model and the right lane change prediction model, and finally the vehicle behavior trend is judged according to the lane keeping probability, left lane change probability and right lane change probability output by the lane keeping prediction model, the left lane change prediction model and the right lane change prediction model. If the probability value output by a certain model is the largest, it means that the model has the highest matching degree with the obtained observed motion parameter data, and the lane change behavior corresponding to the model is the identified driving behavior. In this scheme, three lane-changing behaviors are used for prediction models of hidden Markov HMM, and the lane-changing start Start, lane-changing Middle and lane-changing end End are used as the hidden states in the hidden Markov HMM model. The observed motion parameters in the model can be reduced. In this scheme, only three motion parameters, namely lateral displacement, lateral velocity and vehicle yaw angular velocity, are selected, which can greatly reduce the amount of model calculation and ensure recognition accuracy at the same time. This method can identify the lane-changing intention behavior of drivers of surrounding vehicles in advance, so that corresponding response strategies can be made in advance when the vehicle is driven automatically or by the driver, thereby avoiding accidents. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0030] Figure 1 Flow chart of a vehicle driving behavior recognition method according to an embodiment of the present invention;
[0031] Figure 2 A schematic diagram of the steps of training a model in an embodiment of the present invention;
[0032] Figure 3 Schematic diagram of state transfer in an embodiment of the present invention; DETAILED DESCRIPTION
[0033] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.
[0034] The embodiments of the present invention are further described in detail below with reference to the accompanying drawings.
[0035] Figure 2 Schematic diagram of the steps of training the model in an embodiment of the present invention; Figure 2 As shown, the present invention provides a vehicle driving behavior recognition method, comprising:
[0036] According to the vehicle's driving trajectory information, the vehicle's lateral behavior is classified into three categories: left lane change, right lane change, and lane keeping. By studying the driver's lane change behavior and the vehicle's motion trajectory, it can be found that each behavior can be divided into three states: lane change start, lane change process, and lane change end. In the three states, the change trends of lateral displacement, lateral velocity, and yaw angular velocity are different.
[0037] S0: Training the lane keeping prediction model, left lane change prediction model and right lane change prediction model. Specifically including:
[0038] S01: Establishing an initial lane-changing behavior prediction model for each lane-changing behavior.
[0039] Based on the lane changing state, observed motion parameters, initial state distribution probability, state transition probability, observed motion parameter mean and observed motion parameter variance of each lane changing behavior, three initial lane changing behavior prediction models are established, including the initial lane keeping prediction model, the initial left lane changing prediction model and the right lane changing prediction model.
[0040] In this embodiment, the vehicle is a vehicle participating in traffic around the vehicle, and the lateral behaviors of the vehicles around the vehicle are divided into three categories: left lane change, right lane change, and lane keeping. According to these three categories, in order to identify the lane changing behavior of the current vehicle, it is necessary to know the probability values of the three lane changing behaviors occurring in the current driving trajectory, and the lane changing behavior of the current vehicle is obtained by comparing the probability values. Therefore, based on the Bayesian principle, a hidden Markov model (HMM) for the three lane changing behaviors is established, and three initial lane changing behavior prediction models are established, namely, the initial left lane changing prediction model, the initial lane keeping prediction model, and the initial right lane changing prediction model, corresponding to the initial lane changing behavior prediction model I c The basic parameter set is λ c ={S, O, π, A, μ, U}, where S is the lane change state, O is the observed motion parameter, π is the initial state distribution probability, A is the state transition probability, μ is the mean of the observed motion parameter, and U is the variance of the observed motion parameter.
[0041] Hidden Markov (HMM) model is a type of Bayesian network and is an incomplete data statistical model. For the Markov chain Markov model, the state of the Markov model is visible, while for those problems where the model state is invisible and the observed time series does not correspond one-to-one with the state, the hidden Markov HMM model is needed to solve it. In other words, in the hidden Markov HMM model, the state cannot be directly observed, which is called a hidden state, and only the observation sequence of the model generated based on the state can be seen. The hidden Markov HMM contains two random processes, a Markov chain Markov used to describe the transition between states and an output process that describes the correspondence between the state and the observation sequence. Through these two processes, the model can obtain information related to the state sequence through the observation sequence.
[0042] In some optional embodiments, the lane changing state includes lane changing start, lane changing in progress, and lane changing end. The lane changing start, lane changing in progress, and lane changing end states cannot be directly measured by sensors, so they are set as hidden states S, and the lane changing start, lane changing in progress, and lane changing end states are regarded as hidden states. The number of hidden states in the hidden Markov HMM model is 3.
[0043] In some optional embodiments, the observed motion parameters include the lateral displacement of the vehicle relative to the center line of the lane within a set time period, the lateral velocity and the vehicle yaw rate.
[0044] To a certain extent, the more observation variables are selected, the higher the model recognition rate will be, but it will cause the problem of increased parameter evidence dimension, overly complex model and large amount of calculation. Reasonable selection of observation variables that can fully reflect different driving behaviors can make the model calculation small and the recognition rate high. When the vehicle is in lane keeping behavior and left and right lane changing behavior, the lateral related physical quantities will have a significant change trend, so the above three parameters are selected. The observation quantities of this scheme are the lateral displacement, lateral velocity and yaw rate of the vehicle, and the calculation amount is small. The initial lane keeping prediction model, the initial left lane change prediction model and the right lane change prediction model are established for the three lane changing behaviors respectively, and the lane change start Start, lane change Middle and lane change end End of each model are used as the hidden state in the hidden Markov HMM model, so that each model can ensure the accuracy of each behavior recognition when the observation quantities are only the lateral displacement, lateral velocity and yaw rate of the vehicle, and reduce the difficulty of calculation.
[0045] In some optional embodiments, the set time period is 4S and the interval is 0.05S. Since the identification of the vehicle's behavioral intention requires the change of the observed variables within a time window, the Hidden Markov Model HMM needs to input the observed variable sequence with a time length of the set time period, that is, the observed motion parameters in this example. The observed motion parameters within the set time period T are recorded as: O = [o1, ·, o t ,·,o T ], o t is the sequence of observed variables corresponding to time t, that is, the observed motion parameters at time t within the set time period. t =[d t , v t , w rt ],d t is the lateral displacement of the vehicle relative to the lane centerline in the observed motion parameters at time t, v t is the lateral velocity of the vehicle relative to the center line of the lane in the observed motion parameter at time t, w rt is the vehicle yaw angular velocity at time t.
[0046] In this example, the time period is set to 4S, the interval is 0.05S, and the total number of data of the observed motion parameters within the set time period is 80. When collecting, the observed motion parameters of 0-3.95S are first collected as the first group of data, and then the observed motion parameters are rolled and updated.
[0047] The state transition probability is the probability of a lane-changing state at a certain moment transferring to any lane-changing state at the next moment. Specifically, the state transition probability is the probability that the vehicle transfers from the lane-changing start state at a certain moment to the lane-changing middle state at the next moment, the lane-changing end state, or maintains the current lane-changing start state; the probability that the vehicle transfers from the lane-changing middle state at a certain moment to the lane-changing start state at the next moment, the lane-changing end state, or maintains the current lane-changing middle state; the probability that the vehicle transfers from the lane-changing end state at a certain moment to the lane-changing start state at the next moment, the lane-changing middle state, or maintains the current lane-changing middle state.
[0048] Figure 3 Schematic diagram of state transition in an embodiment of the present invention, such as Figure 3 As shown in the figure, S1, S2 and S3 are respectively the start state, the lane changing middle state and the lane changing end state. The state transition probability A is expressed in the form of a matrix, that is, A = [a ij ] 3×3 , where a ij =P(q t+1 =s j |q t =s i ), 1≤i, j≤3, indicating that the state is q at time t i Under the condition of t plus 1 time interval, that is, transfer to state q at the next moment j The probability of lane change is , i is the current state, j is the next state, si, sj are any of the lane change start, lane change middle and lane change end states, and q is a formula parameter.
[0049] In this embodiment, the observed motion parameter mean μ={μ1, μ2, μ3}, where μ1 is the mean of the lateral displacement of the vehicle relative to the center line of the lane among all the observed motion parameters within the set time period, μ2 is the mean of the lateral velocity of the vehicle relative to the center line of the lane among all the observed motion parameters within the set time period, and μ3 is the mean of the yaw angular velocity of the vehicle among all the observed motion parameters within the set time period.
[0050] The observed motion parameter variance U = {U1, U2, U3}, U1 is the variance of the lateral displacement of the vehicle relative to the center line of the lane among all the observed motion parameters within the set time period, μ2 is the variance of the lateral velocity of the vehicle relative to the center line of the lane among all the observed motion parameters within the set time period, and μ3 is the variance of the yaw angular velocity of the vehicle among all the observed motion parameters within the set time period.
[0051] In some optional embodiments, the initial state distribution probability is set to 0.5, 0.4 and 0.1. In this embodiment, the initial state distribution probability is set based on experience, and the sum of the initial state distribution probabilities is guaranteed to be 1.
[0052] The corresponding initial lane-changing behavior prediction model is trained according to the sample data of each lane-changing behavior to obtain the optimized lane-changing behavior prediction model for each lane-changing behavior. Specifically, it includes:
[0053] S02: Collect the lane-changing behavior of the vehicle in the actual traffic scene and the corresponding observed motion parameters. This step specifically includes:
[0054] The information such as the observed motion trajectory of the vehicle position relative to the lane line and the observed motion parameters of the vehicle in the actual traffic scene is collected, and then the sample data of lane keeping, left lane change and right lane change are filtered out respectively.
[0055] S03: Based on the Baum-Welch algorithm, the observed motion parameters corresponding to each are sequentially introduced into the initial lane changing behavior prediction model corresponding to the lane changing behavior for training, so as to obtain the trained lane keeping prediction model, the left lane change prediction model and the right lane change prediction model.
[0056] In this embodiment, the model parameters are estimated based on the selected sample data of lane keeping, left lane change and right lane change in combination with the Baum-Welch algorithm to complete the parameter training of the established initial lane change behavior prediction model.
[0057] The Baum-Welch algorithm-based method sequentially introduces the observed motion parameters corresponding to each into the initial lane changing behavior prediction model corresponding to the lane changing behavior for training to obtain an optimized lane changing behavior prediction model, specifically including: sequentially introducing the observed motion parameters corresponding to each into the initial lane changing behavior prediction model corresponding to the lane changing behavior to obtain a joint distribution function of the iterative lane changing behavior prediction model, and maximizing the expectation of the joint distribution function to obtain an updated iterative lane changing behavior prediction model, until the state transition probability in the iterative lane changing behavior prediction model converges.
[0058] The model parameters are estimated by combining the Baum-Welch algorithm. The Baum-Welch algorithm uses the principle of the EM algorithm, that is, given the sample observation motion parameter O of each lane-changing behavior, an optimized lane-changing behavior prediction model Iy is determined by calculation to maximize the conditional probability P(O|I), and finally the lane keeping prediction model Iy of the lane keeping behavior is obtained. yk , Left lane change prediction model I yl , Right lane change prediction model I yr .
[0059] Specifically, the observation motion parameters corresponding to each are sequentially introduced into the initial lane-changing behavior prediction model corresponding to the lane-changing behavior for training. The algorithm is in the initial state when the number of iterations h=0, that is, the initial state of the initial lane-changing behavior prediction model mentioned above. (h) is the model parameter at the hth iteration, and λ is the initial lane-changing behavior prediction model parameter set mentioned above.
[0060] Then at the h+1th iteration, the joint distribution function P(O,I;λ) is based on the conditional probability P(O|I;λ (h) ) is expected to be:
[0061] Maximize the above formula, and then get the updated model parameters as follows:
[0062] Iterate continuously until the value of the state transition probability converges. Finally, the parameter training of the established model is completed through the above process, and a set of λ values is obtained, which are the optimized model parameters of the optimized lane-changing behavior prediction model Iy.
[0063] Figure 1 FIG. 1 is a flow chart of a vehicle driving behavior recognition method according to an embodiment of the present invention; Figure 1 As shown, after the above training optimization, the initial left lane change prediction model, the initial lane keeping prediction model and the initial right lane change prediction model are transformed into the trained and optimized left lane change prediction model, the lane keeping prediction model and the right lane change prediction model. When used, the following steps are specifically included:
[0064] S1: Obtain the observed motion parameters of the target vehicle.
[0065] In this embodiment, the observed motion parameters of the vehicle are collected in real time during the driving process of the vehicle, including the lateral displacement of the vehicle relative to the center line of the lane in all the observed motion parameters in each set time period, the lateral speed of the vehicle in all the observed motion parameters in each set time period, and the yaw angular velocity of the vehicle in all the observed motion parameters in each set time period. When collecting, the observed motion parameters of 0-3.95S are first collected as the first group of data, and then the observed motion parameters are rolled and updated as subsequent parameters.
[0066] S2: Inputting the observed motion parameters into a lane keeping prediction model, a left lane change prediction model and a right lane change prediction model respectively.
[0067] In this example, the lateral displacement, lateral velocity and yaw rate of the vehicle relative to the lane centerline among all the observed motion parameters collected within a set time period are input into the keeping prediction model, the left lane change prediction model and the right lane change prediction model.
[0068] S3: Determine the vehicle behavior trend according to the lane keeping probability, left lane changing probability and right lane changing probability output by the lane keeping prediction model, the left lane changing prediction model and the right lane changing prediction model.
[0069] In this example, based on the known parameters of the three optimization models for lane keeping, left lane change, and right lane change and the collected observed motion parameter sequence, the forward-backward strategy in the hidden Markov model HMM algorithm is used to calculate the probability value of a certain observed motion parameter belonging to each driving behavior. Finally, the final behavior of the current vehicle is obtained by comparing the three probability values.
[0070] The observed motion parameters are respectively input into the lane keeping prediction model, the left lane change prediction model and the right lane change prediction model to obtain the lane keeping probability, the left lane change probability and the right lane change probability, specifically including:
[0071] Using the probability density function formula in Gaussian distribution, the probability of hidden state j transferring to visible state can be obtained as:
[0072] For a given driving sequence sample, the forward algorithm based on HMM calculates the probability of the three left lane change prediction models, lane keeping prediction model and right lane change prediction model generating the sequence. If the probability value output by a certain model is the largest, it means that the model has the highest match with the given observation data, and the driving behavior corresponding to the model is the identified driving behavior.
[0073] (1) Initial value
[0074] a1(i)=π i b i (o1), i = 1, 2, ..., N, where N is the number of hidden states, 3 in this example, and α is the forward variable introduced in the algorithm; α t (i) represents the probability that a partial observation sequence is generated before time t in the model and the state is Si at time t, and O1 is the observation motion parameter in the first time period.
[0075] (2) Recursion
[0076] i=1,2,…,N,j=1,2…,N,α t+1 (i) represents the probability that the state is Si at time t+1, and that the partial observation sequence is generated before time t+1 in the model. t+1 is the sequence of observed variables corresponding to the next moment of time t.
[0077] (3) Final output
[0078]
[0079] Optimized Lane Keeping Prediction Model I yk The optimization parameter set is λ k , Optimized left lane change prediction model I yl The optimized parameter set is λ1, and the optimized right lane change prediction model I yr The parameter set is λ r , the corresponding probability value of each lane-changing behavior is calculated as P(O|λ K ), P(O|λ1) and P(O|λ r ). The driving behavior corresponding to the maximum value among the three is the lane-changing behavior TCI of the current vehicle. i , TCI i =max{P(O|λ k ), P(O|λ l ), P(O|λ r )}.
[0080] In summary, the vehicle driving behavior recognition method, based on the lane changing state, observed motion parameters, initial state distribution probability, state transition probability, observed motion parameter mean and observed motion parameter variance of each lane changing behavior, respectively establishes hidden Markov models for three driving behaviors of lane keeping, left lane changing and right lane changing, namely, the initial lane changing behavior prediction model, including the initial left lane changing prediction model, the initial lane keeping prediction model and the initial right lane changing prediction model, and takes the lane changing start Start, lane changing Middle and lane changing end End as the hidden states in the hidden Markov HMM model, which ensures the accuracy of each behavior recognition and makes the behavior recognition more accurate; the corresponding initial lane changing behavior prediction model is trained according to the sample data of each lane changing behavior, and the optimized lane changing behavior prediction model of each lane changing behavior is obtained, including the left lane changing prediction model, the lane keeping prediction model and the right lane changing prediction model.
[0081] When using the optimized left lane change prediction model, lane keeping prediction model and right lane change prediction model, first obtain the observed motion parameters of the target vehicle, the observed motion parameters are the lateral displacement, lateral velocity and vehicle yaw rate of the target vehicle relative to the lane centerline, and input the observed motion parameters into the lane keeping prediction model, the left lane change prediction model and the right lane change prediction model respectively, and finally judge the vehicle behavior trend according to the lane keeping probability, left lane change probability and right lane change probability output by the lane keeping prediction model, the left lane change prediction model and the right lane change prediction model. If the probability value output by a certain model is the largest, it means that the model has the highest matching degree with the obtained observed motion parameter data, and the lane change behavior corresponding to the model is the identified driving behavior. In this scheme, three lane-changing behaviors are used for prediction models of hidden Markov HMM, and the lane-changing start Start, lane-changing Middle and lane-changing end End are used as the hidden states in the hidden Markov HMM model. The observed motion parameters in the model can be reduced. In this scheme, only three motion parameters, namely lateral displacement, lateral velocity and vehicle yaw angular velocity, are selected, which can greatly reduce the amount of model calculation and ensure recognition accuracy at the same time. This method can identify the lane-changing intention behavior of drivers of surrounding vehicles in advance, so that corresponding response strategies can be made in advance when the vehicle is driven automatically or by the driver, thereby avoiding accidents.
[0082] On the other hand, the present invention also provides a vehicle driving behavior recognition device, including: a parameter acquisition module, a calculation and analysis module and a judgment and output module.
[0083] Among them, the parameter acquisition module is used to obtain the observed motion parameters of the target vehicle; the calculation and analysis module is used to input the observed motion parameters into the lane keeping prediction model, the left lane change prediction model and the right lane change prediction model respectively; the judgment output module is used to judge the vehicle behavior trend according to the lane keeping probability, left lane change probability and right lane change probability output by the lane keeping prediction model, the left lane change prediction model and the right lane change prediction model.
[0084] In this embodiment, the observed motion parameters of the vehicle are collected in real time during the driving process of the vehicle through the parameter acquisition module, including the lateral displacement of the vehicle relative to the center line of the lane in all the observed motion parameters in each set time period, the lateral speed of the vehicle in all the observed motion parameters in each set time period, and the yaw angular velocity of the vehicle in all the observed motion parameters in each set time period. Then, the observed motion parameters are input into the lane keeping prediction model, the left lane change prediction model and the right lane change prediction model respectively through the calculation and analysis module for calculation and analysis to obtain the lane keeping probability, the left lane change probability and the right lane change probability. If the probability value output by a certain model is the largest, it means that the model has the highest matching degree with the obtained observed motion parameter data, and the lane changing behavior corresponding to the model is the identified driving behavior. Since three lane-changing behaviors are predicted based on Hidden Markov Model (HMM), and the lane-changing start, lane-changing middle, and lane-changing end are used as the hidden states in the Hidden Markov Model (HMM), the observed motion parameters in the model can be reduced. In this solution, only three motion parameters, namely, lateral displacement, lateral velocity, and vehicle yaw rate, are selected, which can greatly reduce the amount of calculation of the model and ensure recognition accuracy. In addition, the device can identify the lane-changing intention of drivers of surrounding vehicles in advance, so that corresponding response strategies can be made in advance when the vehicle is driven automatically or by the driver, thereby avoiding accidents.
[0085] It should be noted that, in this application, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.
[0086] The above description is only a specific implementation of the present application, so that those skilled in the art can understand or implement the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest range consistent with the principles and novel features applied for herein.
Claims
1. A vehicle driving behavior recognition method, characterized in that: The following steps are involved: The lane keeping prediction model, the left lane change prediction model and the right lane change prediction model are trained. The methods of training the lane keeping prediction model, the left lane change prediction model and the right lane change prediction model all include: Establishing the initial lane-changing behavior prediction model for each lane-changing behavior. The initial lane-changing behavior prediction model is a hidden Markov model for three lane-changing behaviors based on the Bayesian principle, with the lane-changing start, lane-changing, and lane-changing end of each model as the hidden states in the hidden Markov model. Collecting observed motion parameters corresponding to each lane-changing behavior of the vehicle in an actual traffic scenario, wherein the observed motion parameters include the lateral displacement of the vehicle relative to the center line of the lane within a set time period, the lateral velocity, and the yaw angular velocity of the vehicle; Based on the Baum-Welch algorithm, the observed motion parameters corresponding to each are sequentially introduced into the initial lane-changing behavior prediction model corresponding to the lane-changing behavior for training, and the trained lane keeping prediction model, left lane change prediction model and right lane change prediction model are obtained; Obtaining observed motion parameters of the target vehicle; The observed motion parameters are respectively input into the lane keeping prediction model, the left lane change prediction model and the right lane change prediction model to obtain the lane keeping probability, the left lane change probability and the right lane change probability, specifically including: using the probability density function formula in the Gaussian distribution, the probability of the hidden state j transferring to the visible state is obtained as: Where O is the observed motion parameter, π is the initial state distribution probability, μ is the mean of the observed motion parameters, U is the variance of the observed motion parameters, and the vehicle behavior trend is determined according to the lane keeping probability, left lane changing probability and right lane changing probability output by the lane keeping prediction model, the left lane changing prediction model and the right lane changing prediction model; If the probability value output by a certain model is the largest, it means that the model has the highest matching degree with the acquired observed motion parameter data, and the lane changing behavior corresponding to the model is the identified driving behavior.
2. The vehicle driving behavior recognition method according to claim 1, characterized in that: The Baum-Welch algorithm is based on the initial lane changing behavior prediction model corresponding to the lane changing behavior, and the observed motion parameters corresponding to each are sequentially introduced for training to obtain the trained lane keeping prediction model, the left lane change prediction model and the right lane change prediction model, including: Substituting the observed motion parameters corresponding to each into the initial lane-changing behavior prediction model corresponding to the lane-changing behavior in turn, and obtaining the joint distribution function of each iterative lane-changing behavior prediction model; The expectation of each joint distribution function is maximized to obtain an updated iterative lane changing behavior prediction model until the state transition probability in the iterative lane changing behavior prediction model converges. The obtained iterative lane changing behavior prediction models are used as the trained lane keeping prediction model, left lane change prediction model and right lane change prediction model.
3. The vehicle driving behavior recognition method according to claim 1, characterized in that: The training data of the lane keeping prediction model, the left lane change prediction model and the right lane change prediction model all include lane change state, observed motion parameters, initial state distribution probability, state transition probability, observed motion parameter mean and observed motion parameter variance.
4. The vehicle driving behavior recognition method according to claim 3, characterized in that: The lane changing status includes lane changing start, lane changing in progress and lane changing end.
5. The vehicle driving behavior recognition method according to claim 3, characterized in that: The state transition probability is the probability that a certain lane changing state at a certain moment is transferred to any lane changing state at the next moment.
6. The vehicle driving behavior recognition method according to claim 3, characterized in that: The lane changing behaviors include left lane changing, lane keeping and right lane changing.
7. A vehicle driving behavior recognition device, characterized in that: include: A parameter acquisition module, which is used to obtain the observed motion parameters of the target vehicle; A calculation and analysis module is used to input the observed motion parameters into a lane keeping prediction model, a left lane change prediction model, and a right lane change prediction model, respectively. The methods for training the lane keeping prediction model, the left lane change prediction model, and the right lane change prediction model all include: Establishing the initial lane-changing behavior prediction model for each lane-changing behavior. The initial lane-changing behavior prediction model is a hidden Markov model for three lane-changing behaviors based on the Bayesian principle, with the lane-changing start, lane-changing, and lane-changing end of each model as the hidden states in the hidden Markov model. Collecting observed motion parameters corresponding to each lane-changing behavior of the vehicle in an actual traffic scenario, wherein the observed motion parameters include the lateral displacement of the vehicle relative to the center line of the lane within a set time period, the lateral velocity, and the yaw angular velocity of the vehicle; Based on the Baum-Welch algorithm, the observed motion parameters corresponding to each are sequentially introduced into the initial lane-changing behavior prediction model corresponding to the lane-changing behavior for training, and the trained lane keeping prediction model, left lane change prediction model and right lane change prediction model are obtained; The judgment output module is used to input the observed motion parameters into the lane keeping prediction model, the left lane change prediction model and the right lane change prediction model respectively to obtain the lane keeping probability, the left lane change probability and the right lane change probability, specifically including: using the probability density function formula in the Gaussian distribution to obtain the probability of the hidden state j transferring to the visible state: Where O is the observed motion parameter, π is the initial state distribution probability, μ is the mean of the observed motion parameters, U is the variance of the observed motion parameters, and the vehicle behavior trend is judged according to the lane keeping probability, left lane change probability and right lane change probability output by the lane keeping prediction model, the left lane change prediction model and the right lane change prediction model. If the probability value output by a certain model is the largest, it means that the model has the highest matching degree with the obtained observed motion parameter data, and the lane changing behavior corresponding to the model is the identified driving behavior.