CN112927514B - Prediction method and system for motor vehicle yellow light running behavior based on 3D lidar - Google Patents

Abstract

The invention relates to a method and system for predicting the behavior of motor vehicles running a yellow light based on 3D laser radar, comprising: using 3D laser radar to detect and obtain vehicle trajectory data; The vehicle is classified into the lane where the vehicle is located; after receiving the yellow light start signal, the vehicle trajectory data obtained within the first 1.5s of the yellow light time is input into the trajectory prediction model, and the vehicle within the first 1.5s of the yellow light time is analyzed by the trajectory prediction model. The trajectory data is judged, and the vehicle trajectory data 1.5s after the yellow light time is predicted according to the judgment result; the traffic trend of the vehicle after the yellow light ends is judged according to the vehicle trajectory data 1.5s after the yellow light time. The invention can provide a good basis for solving the traffic safety problem of vehicle intrusion during the phase change of the signal lights at the urban signal control intersection, and has the advantages of not relying on the characteristic information of the moving target, accurate, stable and efficient detection, low cost, good adaptability and the like. .

Description

Prediction method and system for vehicle yellow light running behavior based on 3D lidar

technical field

The invention relates to the field of intelligent traffic perception, in particular to a method and system for predicting the behavior of motor vehicles running a yellow light based on 3D laser radar.

Background technique

The ultra-high population density and increasing car ownership in Chinese cities have contributed to the high density of arterial road networks in urban road construction. At intersections in Chinese cities with traffic light control, longer phase transition times are typically used at the end of the green phase, i.e., 3 seconds for a flashing green indication followed by 3 seconds for a yellow indication. Such long phase transition times lead to heterogeneous decisions. As a result, dangerous driving behaviors such as driving at red lights, sudden stops, aggressive passing, and inconsistent decision-making by vehicles in front and behind are more likely to occur at these intersections, which can lead to right-angle and rear-end collisions.

For the above-mentioned traffic safety problems at intersections, the traditional solutions are mostly by promulgating corresponding traffic safety laws and regulations, increasing enforcement efforts, and severely punishing illegal acts such as running red lights, so as to achieve the purpose of reducing the occurrence of such incidents. Punishment methods usually treat the symptoms but not the root causes, and it is difficult to achieve good results. In recent years, due to the interest in the concept of vehicle-road coordination, some scholars have begun to consider the coordinated control between intelligent road facilities and intelligent vehicles. For prediction and early warning, however, this requires a sufficient equipment base, and at this stage, my country's transportation system and road facilities have not yet met the requirements.

Therefore, developing a system with good adaptability that can actively identify these dangerous behaviors can help solve the driving dilemma of drivers, and provide a technical basis for further prevention and control and safety early warning.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide a method and system for predicting the behavior of a motor vehicle running a yellow light based on a 3D laser radar in order to overcome the safety problems existing at the above-mentioned intersection. The system makes full use of the data returned by 3D lidar, and uses the kinematic characteristics and trajectory prediction of the vehicle to realize real-time detection and prediction of vehicles about to enter the intersection during the duration of the yellow light, so as to actively identify and control the signal lights of urban road intersections. During the phase transition, the vehicle intrusion behavior provides strong technical support, the detection is accurate, stable and efficient, the cost is low, and the adaptability is good.

The object of the present invention can be realized through the following technical solutions:

A method for predicting the behavior of motor vehicle running a yellow light based on 3D lidar, comprising the following steps:

Step 1. Use the 3D laser radar installed at the entrance of the urban intersection to perceive the vehicle about to enter the intersection, and obtain the vehicle trajectory data through the 3D laser radar detection;

Step 2. Map the vehicle trajectory data obtained by the 3D lidar detection into the three-dimensional coordinate system within the range of the entryway, and classify the vehicles according to the lanes where the vehicles are located;

Step 3. After receiving the yellow light start signal, input the vehicle trajectory data obtained in the first 1.5s of the yellow light time into the trajectory prediction model, and use the trajectory prediction model to perform the vehicle trajectory data in the first 1.5s of the yellow light time. Judgment, predict the vehicle trajectory data 1.5s after the yellow light time according to the judgment result;

Step 4. According to the vehicle trajectory data 1.5s after the time of the yellow light, determine the traffic trend of the vehicle after the end of the yellow light, that is, whether the vehicle enters the intersection after the red light turns on at the end of the yellow light; if so, output the result; if not , and return to step 1 for the next round of prediction.

Preferably, the three-dimensional coordinate system is established before acquiring the vehicle trajectory data, and the stop line coordinates of the intersection, the coordinates and range of the lane, and the information of the lane are input in the three-dimensional coordinate system.

Preferably, the vehicle trajectory data includes: the ID of the vehicle, the speed of the vehicle, the acceleration of the vehicle, and the distance between the vehicle and the stop line.

Preferably, the yellow light start signal specifically includes the real-time phase information of the intersection, that is, the current phase and the duration of the current phase.

Preferably, the trajectory prediction model in step 3 is established according to historical vehicle trajectory data, and the establishment of the trajectory prediction model includes the following steps:

Step 3.1. Collect historical vehicle trajectory data within the 3s yellow light time to form vehicle trajectory data set A;

Step 3.2. Perform cluster analysis on the vehicle trajectory data set A to obtain the cluster center trajectory data; according to the clustering results, the data of the vehicle trajectory data set A is divided into i categories, and the i categories are used as i trajectory labels, and each Each category corresponds to a track label;

Step 3.3. Divide the vehicle trajectory data set A into a training set B and a test set C, and use the trajectory data of the training set B and the trajectory label corresponding to each trajectory as input to establish a convolutional neural network (CNN) learning model Learning the historical vehicle trajectory data and its corresponding labels;

Step 3.4: Train the model until the test set C is used to test the model, and the test value reaches the expected accuracy rate, then the trajectory prediction model is established.

Preferably, the vehicle trajectory data or the historical vehicle trajectory data is divided into a straight vehicle data set, and/or a left-turn vehicle data set, and/or a right-turn vehicle data set.

Preferably, the step 3 specifically includes the following steps:

The vehicle trajectory data obtained within the first 1.5s of the yellow light time is input into the trajectory prediction model established in advance, and the trajectory prediction model can predict and obtain the trajectory label to which the trajectory data belongs; according to the predicted trajectory label, The trajectory data 1.5s after the yellow light of the cluster center trajectory data (3s) corresponding to the trajectory label is selected, which is the predicted trajectory data of the vehicle 1.5s after the yellow light, and then the yellow light is judged by the trajectory data. The passing trend of vehicles after the end.

Preferably, the 3D lidar detects whether a single vehicle or a queue of vehicles is passing in each lane of the entrance by detecting vehicles at the entrance of the intersection. The prediction in step 3 is divided into single-vehicle traffic prediction and There are two scenarios for vehicle queue traffic prediction;

The specific prediction process for a single vehicle is: after receiving the yellow light start signal, determine the phase to which the yellow light belongs at this time (that is, whether the yellow light at this time is a left turn yellow light, a right turn yellow light or a straight yellow light lights), different phases need to predict and discriminate the vehicles in their corresponding lanes; if it is the yellow light of the straight phase at this time, after receiving the start signal of the yellow light, the system only needs to carry out Prediction and discrimination; if it is the yellow light of the special phase for left turn at this time, after receiving the start signal of the yellow light, it is only necessary to predict and discriminate the vehicles in the left turn lane at this time;

The specific prediction process for the vehicle queue is as follows: when the 3D lidar collects vehicle trajectory data, all vehicles are collected together; when each vehicle is predicted and discriminated for intrusion, it starts from the first vehicle in the vehicle queue. , and then predict and discriminate one by one; when discriminating one by one, if it is judged that a vehicle will not enter the intersection at the end of the yellow light, then all the vehicles in the queue after the vehicle will be judged to be in the yellow light. Do not break into the intersection when the light ends.

Another object of the present invention is to provide a 3D laser radar-based vehicle running yellow light behavior prediction system, which includes the following modules:

The detection module is used to obtain the vehicle trajectory data at the entrance of the urban intersection;

A data processing module for analyzing and processing the acquired vehicle trajectory data;

The data prediction module is used to predict the running trajectory of the vehicle and determine whether the vehicle has entered the intersection;

Signal light information module, used to obtain signal light information.

Compared with the prior art, the method and system for predicting the behavior of a motor vehicle running a yellow light based on the 3D laser radar provided by the present invention have the following beneficial effects:

1. The equipment used in the detection of vehicle trajectory data by the method of the present invention is the fixed 3D laser radar detection equipment on the side of the entrance road, which adopts historical and real-time radar data, and has the advantages of reasonable cost, high accuracy, low calculation requirements, and adaptability to all-weather roads. Compared with other detection methods such as video detection, the processing efficiency of lidar information is higher, so real-time trajectory prediction can be achieved, and Analysis and processing are made in time to achieve accurate, efficient, stable, and all-weather detection and signal control of the driving behavior of vehicles during the phase transition of the intersection.

2. The present invention provides a complete set of vehicle driving trend prediction scheme for the traffic safety problem of vehicle intrusion during the phase change of signal lights at urban signal-controlled intersections. The scheme can stably, accurately and efficiently identify and predict dangerous behaviors during the phase transition of traffic signals. On this basis, early warning and prevention and control of such dangerous behaviors can be carried out to reduce the safety of vehicles during the phase transition of the intersection. hidden dangers, reduce the occurrence of accidents at intersections, and improve the safety level of urban operations.

3. The continuous trajectory of the vehicle can be accurately obtained only according to the data obtained by the 3D lidar, and then the subsequent analysis and prediction can be carried out without relying on the vehicle-end equipment such as high-precision GPS, etc., the required cost is low, and for the existing The traffic environment is more adaptable.

Description of drawings

FIG. 1 is a schematic work flow diagram of a method for predicting the behavior of a motor vehicle running a yellow light based on a 3D laser radar according to an embodiment of the present invention.

FIG. 2 is a flowchart of establishing a trajectory prediction model provided by an embodiment of the present invention.

Detailed ways

The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are some, but not all, embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work shall fall within the protection scope of the present invention.

Example

The invention provides a 3D laser radar-based vehicle yellow light running behavior prediction system. The system includes a detection module, a data processing module, a data prediction module and a signal light information module. The data processing module is used to analyze and process the obtained vehicle trajectory data; the data prediction module is used to predict the running trajectory of the vehicle and determine whether the vehicle has entered the intersection; the signal light information module is used to obtain the signal light information.

Based on a general inventive concept, the present invention also provides a method for predicting the behavior of a motor vehicle running a yellow light based on a 3D laser radar. Real-time prediction of the time it takes for vehicles entering the intersection to reach the stop line from their current position, so as to provide strong technical support for the active identification and prevention of vehicle intrusion behavior during the phase transition of signal lights at urban road intersections. The detection is accurate, stable, efficient, and cost-effective. Low, good adaptability, as shown in Figure 1, the specific steps of the method are as follows:

Step 1. Use the 3D laser radar installed at the entrance of the urban intersection to perceive the vehicle about to enter the intersection, and obtain the vehicle trajectory data through the 3D laser radar detection; specifically, in this embodiment, the vehicle trajectory data includes: Information such as ID, vehicle speed, vehicle acceleration, vehicle distance from the stop line, etc.; vehicle trajectory data or historical vehicle trajectory data are divided into straight vehicle datasets, and/or left-turn vehicle datasets, and/or right-turn vehicles data set;

The 3D lidar can be installed on the roadside of the entrance road or on the gantry and sign poles. It is necessary to ensure a certain installation height, and the height is required to avoid the occlusion of green plants and signs. Then, the 3D lidar detects and perceives vehicles within the range of the entryway, and this function is completed by the detection module of the system.

Step 2. Map the vehicle trajectory data obtained by the 3D lidar detection into the three-dimensional coordinate system within the range of the entrance road, and classify the vehicles according to the lane where the vehicle is located; the three-dimensional coordinate system is established before acquiring the vehicle trajectory data, and in the three-dimensional coordinate system Enter the stop line coordinates of the intersection, the coordinates and range of the lane, and the information of the lane;

By establishing a three-dimensional coordinate system within the range of the entryway, input the coordinates of the stop line, the coordinates and range of the lane, and the information of the lane in advance in the coordinate system. The 3D lidar perceives the vehicle, obtains the coordinates of the vehicle relative to the radar, maps it in the established coordinate system, and classifies the vehicle according to the lane where the vehicle is located. By giving each vehicle ID, time stamp information, and the position of the vehicle moving in a certain time interval, the real-time running speed and acceleration information of the vehicle can be obtained. This function is completed by the data processing module of the system.

Step 3. After receiving the yellow light start signal, input the vehicle trajectory data obtained in the first 1.5s of the yellow light time into the trajectory prediction model, and use the trajectory prediction model to judge the vehicle trajectory data in the first 1.5s of the yellow light time. Predict the vehicle trajectory data 1.5s after the yellow light time according to the judgment result;

In this embodiment, the yellow light start signal specifically includes the real-time phase information of the intersection, that is, the current phase and the duration of the current phase; through this information, the system can choose to predict and judge different types of vehicles;

Step 4. According to the vehicle trajectory data 1.5s after the time of the yellow light, determine the traffic trend of the vehicle after the end of the yellow light, that is, whether the vehicle enters the intersection after the red light turns on at the end of the yellow light; if so, output the result; if not , and return to step 1 for the next round of prediction.

Specifically, in this embodiment, the trajectory prediction model in step 3 is established according to historical vehicle trajectory data. As shown in FIG. 2 , the establishment of the trajectory prediction model includes the following steps:

Step 3.1. Collect historical vehicle trajectory data within the 3s yellow light time to form vehicle trajectory data set A;

Step 3.2. Perform cluster analysis on the vehicle trajectory dataset A (the amount of data should be large enough, K-Means or DBSCAN) to obtain the cluster center trajectory data; according to the clustering results, the data of the vehicle trajectory dataset A is divided into i categories (The specific value of i is based on the result of clustering, and each intersection will be different), this i category is used as i track labels, and each category corresponds to a track label;

Step 3.3. Divide the vehicle trajectory data set A into a training set B and a test set C, use the trajectory data of the training set B and the trajectory label corresponding to each trajectory as input, and establish a convolutional neural network (CNN) learning model for the history. Trajectory data and its corresponding trajectory labels for learning;

Step 3.4: Train the model until the test set C is used to test the model. If the test value reaches the expected accuracy, the trajectory prediction model is established.

During the establishment process, the vehicle trajectory data within 1.5s before the yellow light of the above trajectory data is extracted as the input, and the latter 1.5s trajectory data is used as the result. Each piece of data is given a clear label to establish a supervised learning process.

Based on the above trajectory prediction model, step 3 specifically includes the following steps:

The vehicle trajectory data obtained within the first 1.5s of the yellow light time is input into the trajectory prediction model established in advance. The trajectory prediction model can predict and obtain the trajectory label to which the trajectory data belongs; according to the predicted trajectory label, select the trajectory label. The trajectory data of the cluster center trajectory data (3s) corresponding to the category of the trajectory label 1.5s after the yellow light is the predicted trajectory data of the vehicle 1.5s after the yellow light. The passing trend of vehicles.

Specifically, the 3D lidar detects the vehicles at the entrance of the intersection to know whether a single vehicle or a queue of vehicles is passing in each lane of the entrance at this time. The prediction in step 3 is divided into single-vehicle traffic prediction and vehicle queue traffic prediction. two cases;

The specific prediction process for a single vehicle is: after receiving the yellow light start signal, determine the phase to which the yellow light belongs at this time (that is, whether the yellow light at this time is a left turn yellow light, a right turn yellow light or a straight yellow light lights), different phases need to predict and discriminate the vehicles in their corresponding lanes; if it is the yellow light of the straight phase at this time, after receiving the start signal of the yellow light, the system only needs to carry out Prediction and discrimination; if it is the yellow light of the special phase for left turn at this time, after receiving the start signal of the yellow light, it is only necessary to predict and discriminate the vehicles in the left turn lane at this time;

The specific prediction process for the vehicle queue is as follows: when the 3D lidar collects vehicle trajectory data, all vehicles are collected together; when each vehicle is predicted and discriminated for intrusion, it starts from the first vehicle in the vehicle queue. , predict and discriminate one by one; when discriminating one by one, if it is judged that a vehicle will not enter the intersection at the end of the yellow light, then all the vehicles in the queue after the vehicle will be judged to be at the yellow light. Don't break into the intersection at the end.

The vehicle yellow light running behavior prediction system provided by this embodiment can be subsequently matched with an early warning system, an information announcement system, etc., to remind and warn road users such as other vehicles and pedestrians within the range of the intersection. For example, sound and photoelectric warning facilities can be set up on the roadside. After the system predicts that a vehicle is about to break into the intersection after the green light ends, the warning system can be controlled to issue sound and light warnings to pedestrians and vehicles in the intersection to remind these road users. Avoid early or go carefully through the intersection. Moreover, in the future scenario of connected vehicles and autonomous vehicles, the system can directly issue early warning information to such vehicles to assist decision-making, thereby effectively reducing the potential safety hazards of vehicles during the phase transition of the intersection and reducing accidents at the intersection. occur.

The high-precision 3D lidar detection technology and vehicle trajectory prediction technology provided in this embodiment can help identify vehicle intrusion behaviors during the phase transition of traffic signals, and make real-time predictions for such dangerous behaviors. On this basis, there can be Ways such as signal light adjustment, sound warning, etc. are used to remind vehicles and pedestrians at the intersection to reduce the safety hazards of vehicles during the phase transition of the intersection, reduce the occurrence of accidents at the intersection, and improve the operational safety level of urban roads.

The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited to this. Any person familiar with the technical field can easily think of various equivalents within the technical scope disclosed by the present invention. Modifications or substitutions should be included within the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims (9)

1. A motor vehicle yellow light running behavior prediction method based on a 3D laser radar is characterized by comprising the following steps:

step 1, sensing a vehicle about to enter an intersection by using a 3D laser radar installed at an entrance lane of the urban intersection, and detecting and acquiring vehicle track data through the 3D laser radar;

step 2, mapping vehicle track data obtained by the 3D laser radar detection into a three-dimensional coordinate system within the range of an entrance lane, and classifying the vehicles according to lanes where the vehicles are located;

step 3, after receiving a yellow light starting signal, inputting vehicle track data acquired within the first 1.5s of the yellow light time into a track prediction model, judging the vehicle track data within the first 1.5s of the yellow light time through the track prediction model, and predicting the vehicle track data within 1.5s after the yellow light time according to a judgment result;

step 4, judging the traffic tendency of the vehicle after the yellow light is finished according to the vehicle track data 1.5s after the yellow light time, namely judging whether the vehicle enters the intersection after the yellow light is finished and the red light is lighted; if yes, outputting the result; if not, returning to the step 1 to predict the next turn.

2. The method for predicting the yellow light running behavior of the motor vehicle based on the 3D laser radar as claimed in claim 1, wherein the three-dimensional coordinate system is established before the vehicle track data is acquired, and stop line coordinates of an intersection, coordinates and range of a lane and information of the lane are input into the three-dimensional coordinate system.

3. The method for predicting the yellow light running behavior of a motor vehicle based on a 3D laser radar as claimed in claim 1, wherein the vehicle track data comprises: the ID of the vehicle, the speed of the vehicle, the acceleration of the vehicle, the distance of the vehicle from the stop line.

4. The method for predicting the yellow light running behavior of the motor vehicle based on the 3D laser radar as claimed in claim 1, wherein the yellow light starting signal specifically comprises real-time phase information of an intersection, namely a current phase and a duration of the current phase.

5. The method for predicting the yellow light running behavior of the motor vehicle based on the 3D laser radar as claimed in claim 1, wherein the track prediction model in the step 3 is established according to historical vehicle track data, and the establishment of the track prediction model comprises the following steps:

step 3.1, collecting historical vehicle track data within 3s of yellow light time to form a vehicle track data set A;

step 3.2, carrying out clustering analysis on the vehicle track data set A to obtain clustering center track data; dividing the data of the vehicle track data set A into i types according to the clustering result, taking the i types as i track labels, wherein each type corresponds to one track label;

3.3, dividing the vehicle track data set A into a training set B and a testing set C, taking the track data of the training set B and a track label corresponding to each track as input, and establishing a convolutional neural network learning model to learn the historical vehicle track data and the label corresponding to the historical vehicle track data;

and 3.4, training the model until the model is tested by using the test set C, and if the test value reaches the expected accuracy, finishing the establishment of the track prediction model.

6. The method for predicting the yellow light running behavior of a motor vehicle based on a 3D laser radar as claimed in claim 5, wherein the vehicle track data or historical vehicle track data are divided into a straight-going vehicle data set and/or a left-turning vehicle data set and/or a right-turning vehicle data set.

7. The method for predicting the yellow light running behavior of the motor vehicle based on the 3D laser radar as claimed in claim 6, wherein the step 3 specifically comprises the following steps:

inputting vehicle track data acquired within the first 1.5s of the yellow light time into a track prediction model established in advance, and predicting by the track prediction model to acquire a track label to which the track data belong; according to the predicted track label, selecting the track data 1.5s behind the yellow light of the clustering center track data of the category corresponding to the track label, namely the track data 1.5s behind the yellow light of the predicted vehicle, and then judging the passing trend of the vehicle after the yellow light is finished by the track data.

8. The method for predicting the yellow light running behavior of the motor vehicle based on the 3D laser radar as claimed in claim 4, wherein the 3D laser radar obtains whether the vehicles pass through each lane of an entrance way or pass through a vehicle queue at the moment through detection of the vehicles at the entrance way of the intersection, and the prediction in the step 3 is divided into two situations of single vehicle passing prediction and vehicle queue passing prediction;

the specific prediction process for a single vehicle is as follows: after receiving a yellow light starting signal, judging the phase position of the yellow light at the moment, wherein different phases need to predict and judge the corresponding lane vehicle; if the yellow light is in the straight-going phase, after the starting signal of the yellow light is received, the system only needs to predict and judge the vehicles on the straight-going lane; if the vehicle is a yellow light with a special left-turn phase, after receiving a starting signal of the yellow light, only the vehicle on the left-turn lane needs to be predicted and judged at the moment;

the specific prediction process for the vehicle queue is as follows: when the 3D laser radar collects vehicle track data, all vehicles are collected integrally; when each vehicle is predicted and judged to break in, the vehicle is predicted and judged one by one from the first vehicle in the vehicle queue; when the vehicles are judged one by one, if a certain vehicle is judged not to break into the intersection when the yellow light is finished, all vehicles in the queue behind the vehicle are judged not to break into the intersection when the yellow light is finished.

9. A motor vehicle yellow light running behavior prediction system based on a 3D laser radar is characterized by comprising the following modules:

the detection module is used for acquiring vehicle track data at an entrance road of an urban intersection;

the data processing module is used for analyzing and processing the acquired vehicle track data;

the data prediction module is used for predicting the running track of the vehicle and judging whether the vehicle breaks into the intersection or not;

the signal lamp information module is used for acquiring signal lamp information;

after receiving a yellow light starting signal, the data prediction module inputs vehicle track data acquired within the first 1.5s of the yellow light time into a track prediction model, judges the vehicle track data within the first 1.5s of the yellow light time through the track prediction model, and predicts the vehicle track data within 1.5s after the yellow light time according to a judgment result;

judging the traffic tendency of the vehicle after the yellow light is finished according to the vehicle track data 1.5s after the yellow light time, namely judging whether the vehicle rushes into the intersection after the yellow light finishes the red light;

the establishment of the track prediction model comprises the following steps:

collecting historical vehicle track data within 3s of yellow light time to form a vehicle track data set A;

performing clustering analysis on the vehicle track data set A to obtain clustering center track data; dividing the data of the vehicle track data set A into i types according to the clustering result, taking the i types as i track labels, wherein each type corresponds to one track label;

dividing the vehicle track data set A into a training set B and a testing set C, taking the track data of the training set B and a track label corresponding to each track as input, and establishing a convolutional neural network learning model to learn historical vehicle track data and the corresponding label;

and training the model until the model is tested by using the test set C, and establishing the track prediction model when the test value reaches the expected accuracy.

Images