A prediction method for vehicle lane-changing intention based on composite model

Abstract

The invention discloses a prediction method for vehicle lane-changing intention based on a composite model, comprising: collecting speed information and position information of a vehicle and surrounding vehicles; arranging the vehicle and surrounding vehicles in the same coordinate system, and the trajectory information of all target vehicles constitutes a training set A; establishing three hidden Markov models, namely a left lane-changing model, a lane-keeping model, and a right lane-changing model; inputting the training set A into the three models for preliminary prediction, and outputting the corresponding probabilities respectively; combining the longitudinal distance between the neighboring vehicle and the corresponding target vehicle and the corresponding probability to form a training set B, and then inputting the training set B into a multi-layer perceptron model, using the real lane-changing intention corresponding to the target vehicle as a label, completing the training of the multi-layer perceptron model, and obtaining a training model; collecting the speed information and position information of the target vehicle and the neighboring vehicle in real time, and inputting them into the training model for predicting the vehicle lane-changing intention, thereby outputting the lane-changing intention of the target vehicle.

Description

Technical Field

[0001] The present invention belongs to the technical field of autonomous driving, and in particular relates to a method for predicting vehicle lane-changing intention based on a composite model. Background Art

[0002] Smart cars are the future direction of transportation, and they have great potential in improving driving safety and traffic efficiency. The decision-making module of smart cars receives environmental information obtained by the external perception system, and predicts the movement of surrounding vehicles and obstacles in the environment, thereby determining the next behavior of the vehicle. Accurately predicting the intentions and trajectories of surrounding vehicles is crucial because it is a guarantee for effectively avoiding collisions and increasing driving safety in complex traffic environments. It can also ensure that the decisions made are safe and in line with social norms, thereby ensuring driving safety and comfort without being too conservative, and avoiding traffic accidents.

[0003] At present, the research on the prediction of lane-changing intention of surrounding vehicles does not consider the impact of the interaction between vehicles on the vehicle's lane-changing intention, but predicts the lane-changing intention of the target vehicle as an independent individual. However, in actual driving situations, the neighboring vehicles of the target vehicle will greatly affect its lane-changing behavior.

[0004] Therefore, how to consider the impact of interactions between vehicles on the vehicle's lane-changing intention, accurately predict the lane-changing intention of surrounding vehicles, and improve the safety of autonomous driving vehicles are technical problems that technicians in this field need to solve. Summary of the invention

[0005] In view of the above-mentioned deficiencies in the prior art, the purpose of the present invention is to provide a method for predicting vehicle lane-changing intention based on a composite model, which can effectively consider the impact of interactions between vehicles on the vehicle's lane-changing intention, accurately predict the lane-changing intention of surrounding vehicles, and improve the safety of autonomous driving vehicles.

[0006] The technical solution of the present invention is achieved in this way:

[0007] A method for predicting vehicle lane-changing intention based on a composite model comprises the following steps:

[0008] (1) Collect speed and location information of the vehicle and surrounding vehicles;

[0009] (2) Arranging the vehicle and surrounding vehicles in the same coordinate system according to the speed information and position information collected in step (1) to obtain the trajectory information of the vehicle and surrounding vehicles, where the surrounding vehicles immediately adjacent to the vehicle are target vehicles, and the vehicles surrounding each target vehicle are neighbor vehicles of the target vehicle. The trajectory information of all target vehicles constitutes the training set A;

[0010] (3) Establish three hidden Markov models, namely left lane change model, lane keeping model, and right lane change model;

[0011] (4) Input the training set A in step (2) into the left lane change model, lane keeping model and right lane change model for preliminary prediction, and output the corresponding probability P respectively. L , P K , P R ;

[0012] (5) The longitudinal distance between the neighbor vehicle and the corresponding target vehicle and the corresponding probability P L , P K , P R Combine to form a training set B, then input the training set B into the multi-layer perceptron model, use the real lane-changing intention corresponding to the target vehicle as a label, complete the training of the multi-layer perceptron model, and obtain a training model;

[0013] (6) The speed and position information of the target vehicle and neighboring vehicles are collected in real time and input into the training model in step (5) to predict the lane-changing intention of the vehicle, thereby outputting the lane-changing intention of the target vehicle.

[0014] Furthermore, in step (1), the speed information and position information of the vehicle and surrounding vehicles are collected by using a radar device and a camera device on the vehicle.

[0015] Furthermore, the radar device includes but is not limited to millimeter wave radar, laser radar and ultrasonic radar.

[0016] Furthermore, the camera device includes but is not limited to a surround-view camera and a front-view camera.

[0017] Furthermore, the speed information and position information include but are not limited to the vehicle speed, the vehicle heading angle, the speed of surrounding vehicles, lane line information, and the lateral distance and longitudinal distance between the surrounding vehicles and the vehicle.

[0018] Furthermore, the neighbor vehicles are vehicles located in front of and behind the lane where the target vehicle is located and within a set distance of the left lane and the right lane.

[0019] Compared with the prior art, the present invention has the following beneficial effects:

[0020] The present invention first uses the time series modeling capability of the hidden Markov model for the lateral displacement of the target vehicle, the longitudinal speed and other continuous observation variables to preliminarily identify the vehicle's lane-changing intention, and then combines the probability output by the LCL-HMM, LK-HMM and LCR-HMM models with the longitudinal distance between the neighboring vehicle and the corresponding target vehicle, and inputs them into the multi-layer perceptron model to achieve effective prediction of the lane-changing intention of the surrounding target vehicles. In this way, the interaction between the target vehicle and its corresponding neighboring vehicle is used as a factor to predict the target vehicle's lane-changing intention, which can effectively improve the accuracy of the target vehicle's lane-changing intention prediction and the safety of the autonomous driving vehicle. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 -Modeling flow chart of the present invention.

[0022] Figure 2 -Scene diagram of the actual vehicle layout in the embodiment. DETAILED DESCRIPTION

[0023] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0024] A method for predicting vehicle lane-changing intention based on a composite model comprises the following steps:

[0025] (1) Collect speed and location information of the vehicle and surrounding vehicles;

[0026] (2) Arranging the vehicle and surrounding vehicles in the same coordinate system according to the speed information and position information collected in step (1) to obtain the trajectory information of the vehicle and surrounding vehicles, where the surrounding vehicles immediately adjacent to the vehicle are target vehicles, and the vehicles surrounding each target vehicle are neighbor vehicles of the target vehicle. The trajectory information of all target vehicles constitutes the training set A;

[0027] (3) Establish three hidden Markov models, namely the left lane change model LCL-HMM, the lane keeping model LK-HMM, and the right lane change model LCR-HMM;

[0028] (4) Input the training set A in step (2) into the three models of LCL-HMM, LK-HMM and LCR-HMM for preliminary prediction, and output the corresponding probability P respectively. L , P K , P R ;

[0029] (5) The longitudinal distance between the neighbor vehicle and the corresponding target vehicle and the corresponding probability P L , P K , P RCombine to form a training set B, then input the training set B into a multi-layer perceptron (MLP) model, use the actual lane-changing intention corresponding to the target vehicle as a label, complete the training of the multi-layer perceptron model, and thus establish a training model, namely, the HMM-MLP model;

[0030] (6) The speed and position information of the target vehicle and neighboring vehicles are collected in real time and input into the training model in step (5) to predict the lane-changing intention of the vehicle, thereby outputting the lane-changing intention of the target vehicle.

[0031] The flow chart of establishing the HMM-MLP model is as follows Figure 1 As shown. The hidden Markov model (HMM) is first used to model the time series of continuous observation variables such as the lateral displacement and longitudinal speed of the target vehicle to preliminarily identify the vehicle's lane-changing intention. Then, the likelihood probability output by the LCL-HMM model, LK-HMM model, and LCR-HMM model is combined with the longitudinal distance between the neighboring vehicle and the corresponding target vehicle, and input into the multi-layer perceptron (MLP) model to achieve effective prediction of the lane-changing intention of the surrounding target vehicles. In this way, the interaction between the target vehicle and its corresponding neighboring vehicle is used as a factor to predict the target vehicle's lane-changing intention, which can effectively improve the accuracy of the target vehicle's lane-changing intention prediction and the safety of the autonomous driving vehicle.

[0032] The left lane change model LCL-HMM, the lane keeping model LK-HMM, and the right lane change model LCR-HMM are three independent models, so the probability P predicted by the three models is L , P K , P R There is no necessary connection between them.

[0033] The longitudinal distance P between the neighbor vehicle in the left lane of the target vehicle and the target vehicle in step (5) L Combined with and , the longitudinal distance P between the neighbor vehicle in the left lane of the target vehicle and the target vehicle K Combined with and , the longitudinal distance P between the neighbor vehicle in the left lane of the target vehicle and the target vehicle R and are combined to form the training set B.

[0034] In the specific implementation, in step (1), the radar device and the camera device on the vehicle are used to collect the speed information and position information of the vehicle and the surrounding vehicles.

[0035] In a specific implementation, the radar device includes but is not limited to millimeter wave radar, laser radar and ultrasonic radar.

[0036] In a specific implementation, the camera device includes but is not limited to a surround view camera and a front view camera.

[0037] In specific implementation, the speed information and position information include but are not limited to the vehicle speed, the vehicle heading angle, the speed of surrounding vehicles, lane line information, and the lateral distance and longitudinal distance between the surrounding vehicles and the vehicle.

[0038] In a specific implementation, the neighbor vehicles are vehicles located in front of and behind the lane where the target vehicle is located and within a set distance of the left lane and the right lane.

[0039] The setting distance here is set according to needs, and can generally be set to 120m or 90m.

[0040] In step (5), if there is no vehicle in one of the lanes of the target vehicle or the distance from the target vehicle is greater than the set distance, then to ensure the uniformity of the format, the longitudinal distance between the neighboring vehicle in the lane and the target vehicle is considered to be the set distance.

[0041] See also Figure 2 , car 1 is the target vehicle, cars 2 to 5 are the neighboring vehicles of car 1, of which cars 2 and 3 are in the left lane, car 4 is in the own lane, and car 5 is in the right lane. First, the information data of car 1 is input into the left lane change model LCL-HMM, the lane keeping model LK-HMM, and the right lane change model LCR-HMM for preliminary prediction, and the left lane change probability P is obtained respectively. L , Lane keeping probability P K and the probability of right lane change P R , then P L Combined with the longitudinal distances of car 2 and car 3 to car 1, P K Combined with the longitudinal distance from car 4 to car 1, P R Combined with the longitudinal distance from car 5 to car 1, it is then input into the multi-layer perceptron model for training. In this way, the HMM-MLP model is established. After the HMM-MLP model is established, the collected data is directly input into the model to obtain the prediction result.

[0042] Finally, it should be noted that the above embodiments of the present invention are only examples for illustrating the present invention, and are not intended to limit the implementation methods of the present invention. For those skilled in the art, other different forms of changes and modifications can be made based on the above description. It is impossible to list all implementation methods here. Any obvious changes or modifications derived from the technical solution of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for predicting vehicle lane-changing intention based on a composite model, characterized in that: The following steps are involved: (1) Collect speed and location information of the vehicle and surrounding vehicles; (2) Arranging the vehicle and surrounding vehicles in the same coordinate system according to the speed information and position information collected in step (1) to obtain the trajectory information of the vehicle and surrounding vehicles, where the surrounding vehicles immediately adjacent to the vehicle are target vehicles, and the vehicles surrounding each target vehicle are neighbor vehicles of the target vehicle. The trajectory information of all target vehicles constitutes the training set A; (3) Establish three hidden Markov models, namely left lane change model, lane keeping model, and right lane change model; (4) Input the training set A in step (2) into the left lane change model, lane keeping model and right lane change model for preliminary prediction, and output the corresponding probability P respectively. L , P K , P R ; (5) The longitudinal distance between the neighbor vehicle and the corresponding target vehicle and the corresponding probability P L , P K , P R Combine to form a training set B, then input the training set B into the multi-layer perceptron model, use the real lane-changing intention corresponding to the target vehicle as a label, complete the training of the multi-layer perceptron model, and obtain a training model; (6) The speed and position information of the target vehicle and neighboring vehicles are collected in real time and input into the training model in step (5) to predict the lane-changing intention of the vehicle, thereby outputting the lane-changing intention of the target vehicle.

2. The method for predicting vehicle lane-changing intention based on a composite model according to claim 1, characterized in that: In step (1), the radar device and the camera device on the vehicle are used to collect the speed information and position information of the vehicle and surrounding vehicles.

3. The method for predicting vehicle lane-changing intention based on a composite model according to claim 2, characterized in that: The radar device is a millimeter wave radar, a laser radar or an ultrasonic radar.

4. The method for predicting vehicle lane-changing intention based on a composite model according to claim 2, characterized in that: The camera device is a surround view camera or a front view camera.

5. The method for predicting vehicle lane-changing intention based on a composite model according to claim 2, characterized in that: The speed information and position information include the vehicle speed, the vehicle heading angle, the speed of surrounding vehicles, lane line information, and the lateral and longitudinal distances between the surrounding vehicles and the vehicle.

6. The method for predicting vehicle lane-changing intention based on a composite model according to claim 1, characterized in that: The neighbor vehicles are vehicles located in front of and behind the lane where the target vehicle is located, and within a set distance of the left lane and the right lane.

Images