A differential automatic lane-changing decision method

Abstract

The application is suitable for the technical field of auxiliary driving system, and provides a difference automatic lane changing decision method, which comprises the following steps: recording parameters for representing the style of a driver in the driving process, and classifying the driving style by using the parameters; predicting the next track of a surrounding vehicle by using an LSTM neural network based on the observed track of the surrounding vehicle and in combination with the driving style label; selecting a safety distance at the beginning of lane changing as a standard, and presetting different reaction times, maximum accelerations and brake decelerations for different driving styles; selecting a safety related area for the traffic participants around the ego vehicle, and preferentially processing the safety related target when the target exists in the safety related area; and confirming that the lane changing action can be executed when the ego vehicle is in a safety state. The method can significantly improve the safety of the lane changing process, provides a more personalized lane changing mode, and effectively improves the acceptance of the algorithm by the driver and the riding comfort.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of auxiliary driving systems, and particularly relates to a differential automatic lane-changing decision method. BACKGROUND

[0002] In recent years, the auxiliary driving system has developed rapidly, and relevant applications effectively improve the safety during driving and reduce the fatigue of drivers during the journey. In actual driving, lane changing mainly relies on the experience of drivers. Research shows that the probability of traffic accidents caused by lane-changing decision lane selection errors of drivers is as high as 75%, and the number of accidents caused by improper lane changing of drivers accounts for about 6% of the total number of accidents, and the traffic delay caused by the accidents accounts for 10% of the total delay time caused by traffic accidents. Using an autonomous lane-changing system can effectively reduce the probability of accidents during lane changing. In the face of a vehicle changing lanes in front, different drivers have very different reactions. In order to be able to appropriately predict different drivers, it is necessary to classify the driving styles according to the historical driving data of the drivers, so that when predicting the following vehicle, the driving style label and kinematic information can be combined to complete the prediction and achieve a more reasonable decision. For example, for aggressive drivers, a relatively conservative lane-changing decision can be made.

[0003] At present, the data-driven methods for driving behavior analysis and prediction mainly include Hidden Markov Model (HMM), Support Vector Machine (SVM), and Long Short-Term Memory Neural Network (LSTM). Through the analysis and learning of the kinematic data of the target vehicle and surrounding vehicles during driving, a model for judging driving behavior based on the kinematic history data of the vehicle and other vehicle or road attributes can be trained. The traditional lane-changing decision safety discrimination is performed by calculating the time to collision (TTC), but this method considers fewer factors, which may lead to unsafe following conditions. The use of the Responsibility Sensitive Safety Model (RSS) can more comprehensively consider the factors affecting safety and obtain a more reasonable safe following distance.

[0004] In summary, the analysis and prediction of the behaviors of other vehicles during lane changing can help drivers make reasonable judgments and reduce accidents. At the same time, in the automatic driving algorithm, the introduction of the prediction of other vehicles can better make decisions. In terms of lane-changing decision, if only the possibility of collision is judged, it does not conform to the habits of human drivers, and a more reasonable safety distance model is needed to solve this problem. SUMMARY

[0005] The purpose of the embodiments of the application is to provide a differential automatic lane-changing decision method, which aims to solve the problems raised in the background.

[0006] The embodiment of the application is implemented in the following manner: a differential automatic lane-changing decision method, which makes differential automatic lane-changing decision by labeling the driving styles of neighboring vehicles, comprises the following steps:

[0007] Step 1: When the vehicle is running, a data recording system is used to record parameters that can represent the driving style of the driver during driving, and the collected parameters include six items: velocity (Velocity), absolute value of acceleration (AVA), absolute value of jerk (AVJ), throttle pedal opening (TPO), absolute value of throttle pedal opening rate (AVCRTPO) and brake pedal pressure (BPP).

[0008] The mean (Mean), standard deviation (Std), maximum value (Max), minimum value (Min) and median (Median) of these parameters per second are calculated, thereby generating a 30-dimensional feature parameter per second, and the above parameters are used to classify the driving style and form a driving style label.

[0009] Step 2: Based on the observed trajectory of the surrounding vehicle, the driving style label is combined to predict the next trajectory of the vehicle using an LSTM neural network.

[0010] Step 3: When making safety judgment for lane-changing decision, the safety distance at the beginning of lane-changing is selected as the standard. For different driving styles, different reaction times, maximum accelerations and brake decelerations can be preset to increase the practicability of lane-changing safety verification. At the same time, based on V2V communication, the driving style information classified by the vehicle can be transmitted between vehicles.

[0011] Step 4: For the traffic participants around the ego vehicle, a safety-related area can be selected, that is, when there is a vehicle in the target lane within the area, it is preliminarily judged that the lane-changing action is unsafe, and the distances in the area are as follows:

[0012] 1. Longitudinal vehicle driving in the same direction, 2. Vehicle behind the ego vehicle, 3. Opposite approaching vehicle, 4. Lateral expansion of safety-related area.

[0013] Step 5: When there is a target in the safety-related area, the intelligent vehicle should prioritize processing the safety-related target to ensure that it is not in a dangerous state, and when the ego vehicle is in a safe state and the vehicle on the target lane also maintains a safe distance from the ego vehicle, it is confirmed that the lane-changing action can be performed. The selection of the safety-related area ensures that the vehicle leaves sufficient buffer space during lane-changing to avoid reckless lane-cutting behavior, thereby realizing safety monitoring of the autonomous lane-changing algorithm.

[0014] Further technical solutions, items 1 and 2 are longitudinal safety distances, and the calculation method is as follows:

[0015]

[0016] where ρ represents the minimum reaction time of the driver, α max represents the maximum acceleration of the vehicle, β min represents the minimum deceleration when braking, β max represents the maximum deceleration when braking, v r is the current speed of the rear vehicle, v f is the current speed of the front vehicle, and the last value is positive. When the distance is less than this distance, it indicates that the rear vehicle may have a collision risk if no braking measures are taken, so the longitudinal target within this distance is considered as a safety-related target.

[0017] In a further technical solution, the third distance is mainly determined by the safety distance of the oncoming vehicle, and the calculation method is as follows:

[0018]

[0019] where v1 and v2 are the speeds of the ego vehicle and the oncoming vehicle, β min,correct represents the deceleration of the oncoming vehicle, and we believe that β min,correct ≤β min because the vehicle in normal driving is unlikely to consider the oncoming vehicle crossing the line too much. When the distance is less than this distance, we believe that there is a safety impact between the oncoming vehicles, and it should be considered as a safety-related target.

[0020] In a further technical solution, the left and right boundaries of the fourth item are the lateral safety distance, and the calculation method is as follows:

[0021]

[0022] where μ is the safety boundary value of lateral movement, vi represents the current lateral speed of vehicle i, v i,ρ is the lateral speed of vehicle i after the reaction time, and within the reaction time, vehicle i is assumed to perform lateral acceleration movement at α max β min is the minimum lateral deceleration after the reaction time. This distance represents the safety distance of lateral movement, and the target within this lateral range can be considered as a safety-related target.

[0023] In a further technical solution, in the step 1, the K-means clustering method in unsupervised learning is used when classifying data.

[0024] The embodiment of the application provides a differential automatic lane changing decision method, uses adjacent vehicle driving big data, establishes adjacent vehicle driving style labels based on a K-Means clustering analysis method, thereby probabilistically inferring the adjacent vehicle driving style, and provides reliable theoretical support for reasonable lane changing of the vehicle. Meanwhile, the LSTM and the RSS model are fused, the trajectory prediction under a dynamic time window and the lateral and longitudinal safety distance are considered, and more safe technical support is provided for reasonable lane changing of the vehicle. The safety of the lane changing process can be significantly improved, dangerous situations caused by insufficient observation in traditional driving are avoided, inexperienced drivers can be helped to complete lane changing, a more personalized lane changing mode conforming to the personal driving habits of the driver is provided, and the acceptance of the algorithm and the ride comfort of the driver are effectively improved. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 A design flowchart of the differential automatic lane changing decision method provided by the embodiment of the application is provided.

[0026] Figure 2 A flowchart of selecting parameters according to style differences in the differential automatic lane changing decision method provided by the embodiment of the application is provided.

[0027] Figure 3 A safety-related area schematic diagram in the differential automatic lane changing decision method provided by the embodiment of the application is provided.

[0028] Figure 4 An LSTM model diagram in the differential automatic lane changing decision method provided by the embodiment of the application is provided. DETAILED DESCRIPTION

[0029] In order to make the purpose, technical scheme and advantages of the application clearer, the application is further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the application and do not limit the application.

[0030] The specific implementation of the application is described in detail below with reference to specific embodiments.

[0031] The embodiment of the application provides a differential automatic lane changing decision method, which differentiates and automatically makes lane changing decisions by labeling the driving styles of adjacent vehicles, including the following steps:

[0032] Step 1, when the vehicle is driving, a data recording system is used to record the main parameters as shown in Table 1,

[0033] Table 1 main parameters and characteristics

[0034]

[0035] These parameters can characterize the driver's style during driving, and the collected parameters include velocity, absolute value of acceleration (AVA), absolute value of jerk (AVJ), throttle pedal opening (TPO), absolute value of throttle pedal opening rate (AVCRTPO), brake pedal pressure (BPP), a total of six. The average value (Mean), standard deviation (Std), maximum value (Max), minimum value (Min), and median value (Median) of these parameters per second are calculated, thereby generating a 30-dimensional feature parameter per second. The above parameters are used to classify the driving style, forming a driving style label.

[0036] Step 2, based on the observed trajectory of the surrounding vehicle, combined with the driving style label, the next trajectory of the vehicle is predicted using an LSTM neural network. LSTM, which stands for Long Short Term Memory, is a special type of recurrent neural network suitable for processing time series data, and has achieved good results in trajectory prediction. The LSTM unit contains a forgetting gate, a memory gate, and an output, which can be used to simulate the way people remember time series data. Through training, the model can allocate different memory degrees to different time periods of data, achieving good prediction results.

[0037] Step 3, when making safety judgments for lane changing decisions, the safety distance at the start of lane changing is selected as the standard. For different driving styles, different reaction times, maximum accelerations, and brake decelerations can be preset, such as Figure 2 shown, increasing the practicality of lane changing safety verification. At the same time, based on V2V communication, vehicles can exchange driving style information classified by the ego vehicle.

[0038] For traffic participants around the ego vehicle, a safety-related area can be selected, that is, when there is a vehicle in the target lane within the area, it is preliminarily determined that the lane changing action is unsafe, as Figure 3 shown, where the distances are as follows:

[0039] 1: longitudinally forward vehicle driving in the same direction, 2: vehicle behind the ego vehicle, 3: opposite approaching vehicle, 4: lateral expansion of safety-related area.

[0040] The first and second items are longitudinal safety distances:

[0041]

[0042] where ρ represents the minimum reaction time of the driver, α max represents the maximum acceleration of the vehicle, β min represents the minimum deceleration when braking, β max represents the maximum deceleration when braking, v r is the current speed of the rear vehicle, vf is the current speed of the front vehicle, and the last value is positive. When the distance is less than this distance, it means that the rear vehicle may have a collision risk if no braking measures are taken, so the longitudinal target within this distance is considered as a safety-related target.

[0043] The third distance is mainly determined by the oncoming vehicle safety distance, and the calculation method is as follows:

[0044]

[0045] where v1 and v2 are the speeds of the ego vehicle and the oncoming vehicle, β min,correct represents the deceleration of the oncoming vehicle, and we believe that β min,correct ≤β min Because the vehicle in normal driving is likely to consider the oncoming vehicle crossing the line too much. When the distance is less than this distance, we believe that there is a safety impact between the oncoming vehicles, which should be considered as a safety-related target.

[0046] The left and right boundaries of the fourth item are the lateral safety distance, and the calculation method is as follows:

[0047]

[0048] where μ is the safety boundary value of lateral movement, vi represents the current lateral speed of vehicle i, v i,ρ is the lateral speed of vehicle i after the reaction time, and in the reaction time, vehicle i is assumed to perform lateral acceleration movement at α max , β min is the minimum lateral deceleration after the reaction time. This distance represents the safety distance of lateral movement, and the target within this lateral range can be considered as a safety-related target.

[0049] When there is a target in the safety-related area, the intelligent vehicle should preferentially process the safety-related target to ensure that it is not in a dangerous state, and when the ego vehicle is in a safe state and the vehicle on the target lane also maintains a safe distance from the ego vehicle, it is confirmed that the lane changing action can be performed. The selection of the safety-related area ensures that the vehicle leaves sufficient buffer space in the lane changing, avoiding the behavior of reckless lane cutting, and thus realizes the safety monitoring of the autonomous lane changing algorithm.

[0050] As a preferred embodiment of the present application, in the step 1, the K-means clustering method is used when classifying the data. The K-means clustering method in unsupervised learning can show the advantages of simplicity and efficiency in processing large data set tasks. In addition, it has the characteristics of low time complexity, low space complexity and fast convergence speed. Therefore, the K-means algorithm is used to cluster the driving styles. The K-means algorithm is also called K-means clustering algorithm, in which K represents the number of clustering categories, and means represents taking the mean value of the data in each cluster as the center of the cluster, or the centroid, that is, selecting the centroid of each class to describe the cluster.

[0051] Specifically, K initial clustering centers are selected, the distances of each data point to each clustering center are calculated, and the data points are divided into the class where the nearest clustering center is located according to the result. After one classification, the initial clustering centers are updated. The clustering centers are determined by the average value method, and the data points are iteratively assigned until the clustering centers no longer change, which indicates that the clustering criterion function has converged, that is, the clustering analysis process is completed. The final clustering result is to minimize the clustering sum of squares of each class until it is constant. The criterion function is calculated as follows:

[0052]

[0053] In the formula, k is a sample cluster set, n is the number of samples, x i is a sample point, and u j is a sample cluster center.

[0054] The main steps of the K-means clustering method are as follows:

[0055] (1) Select the number of categories K to be clustered;

[0056] (2) Randomly select K data points as clustering centers;

[0057] (3) Each sample point in the data set needs to calculate its Euler distance with each clustering center, so as to find the nearest clustering center point and be assigned to the cluster to which the nearest clustering center point belongs;

[0058] (4) According to the result of the last sample point assignment, the cluster center point is recalculated, and each clustering center is updated;

[0059] (5) Calculate the clustering sum of all points in the data set and the new clustering center point;

[0060] (6) Repeat steps 3-5 above until each data point does not change its own cluster or reaches the set number of iterations.

[0061] The above merely describes preferred embodiments of the present application, and is not used to limit the present application, any modification, equivalent replacement and improvement within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method for differentiating automatic lane-changing decision by labeling the driving style of neighboring vehicles, characterized in that, The method comprises the following steps: Step 1: During the driving of the vehicle, a data recording system is used to record parameters for characterizing the driving style of the driver during driving, the collected parameters including six items of vehicle speed, acceleration absolute value, impact absolute value, throttle pedal opening, throttle pedal opening rate absolute value and brake pedal pressure, the average value, standard deviation, maximum value, minimum value and median value of the parameters in each second are calculated, thereby generating a 30-dimensional feature parameter per second, the driving style is classified using the above parameters, and a driving style label is formed; Step 2: Based on the observed trajectory of the surrounding vehicle, the driving style label is combined, and an LSTM neural network is used to predict the next trajectory; Step 3: When making a safety judgment on the lane changing decision, the safety distance at the beginning of the lane changing is selected as the standard, and different reaction times, maximum accelerations and brake decelerations are preset for different driving styles to ensure the practicability of the lane changing safety verification; At the same time, based on V2V communication, the driving style information classified by the ego vehicle is transmitted between vehicles; Step 4: For the traffic participants around the ego vehicle, a safety-related area is selected, when there is a vehicle in the target lane within the area, it is preliminarily judged that the lane changing action is unsafe, and the following distances are determined: Item 1: longitudinally forward vehicle driving in the same direction, Item 2: vehicle behind the ego vehicle, Item 3: approaching vehicle in the opposite direction, Item 4: safety-related area in the lateral direction; Step 5: When there is a target in the safety-related area, the intelligent vehicle should preferentially process the safety-related target to ensure that it is not in a dangerous state, when the ego vehicle is in a safe state and the vehicle on the target lane also maintains a safe distance from the ego vehicle, it is confirmed that the lane changing action can be performed, the selection of the safety-related area ensures that the vehicle leaves sufficient buffer space during lane changing, and the autonomous lane changing algorithm is safety monitored; In step 4, items 1 and 2 are longitudinal safety distances, and the calculation method is as follows: ； wherein p represents the minimum reaction time of the driver, a max represents the maximum acceleration of the vehicle, b min represents the minimum deceleration when braking, b max represents the maximum deceleration when braking, v r is the current speed of the rear vehicle, v f is the current speed of the front vehicle, the last calculated value being taken as positive; When the vehicle distance is less than this distance, it indicates that the rear vehicle will have a risk of collision if no braking measures are taken, therefore, the longitudinal target within this distance is regarded as a safety-related target; Item 3 distance is determined by the opposite vehicle safety distance, and the calculation method is as follows: ； where v1 and v2 are the speeds of the ego and oncoming vehicles, β min,correct represents the oncoming vehicle deceleration, and β min,correct ≤ β min ; When the vehicle distance is less than this distance, it is determined that the opposite vehicles have a safety impact between them, and they are identified as safety-related targets.

2. The method of claim 1, wherein, The left and right boundaries of item 4 are lateral safety distances, and the calculation method is as follows: ； where µ is the lateral motion safety margin, vi represents the current lateral speed of vehicle i, v i,ρ is the lateral speed of vehicle i after the reaction time, during which vehicle i is assumed to perform a lateral acceleration motion with an acceleration of a max , and β min is the minimum lateral deceleration after the reaction time. The distance represents the safety distance of lateral movement, and the target within the lateral range is regarded as a safety-related target.

3. The method of claim 1, wherein, In step 1, when classifying data, the K-means clustering method in unsupervised learning is adopted.

Images