CN118349694B - Method and system for generating ramp converging region vehicle track database - Google Patents

Abstract

The present invention provides a method and system for generating a vehicle trajectory database in a ramp merging area, belonging to the field of traffic control technology. The method of the present invention includes the following processes: obtaining laser radar point cloud data in the ramp merging area; identifying all vehicle targets according to the laser radar point cloud data; tracking the trajectories of all identified vehicle targets respectively, calculating the driving data of each vehicle target according to the trajectory tracking results of each vehicle target, and generating a vehicle trajectory database according to the trajectory tracking results and driving data of each vehicle target; the present invention selects and utilizes the roadside point cloud data of the ramp merging area, and realizes the construction of a comprehensive and accurate vehicle trajectory database through vehicle target detection and trajectory tracking processing.

Description

Method and system for generating vehicle trajectory database in ramp merging area

Technical Field

The present invention relates to the technical field of traffic control, and in particular to a method and system for generating a vehicle trajectory database in a ramp merging area.

Background Art

Traffic information is an important basic information for urban traffic planning and traffic management. By obtaining comprehensive, rich and real-time traffic information, we can not only grasp the current status of urban road traffic capacity, but also give real-time warnings for potential traffic conflicts, maximize the road capacity itself, and ensure traffic safety. Traffic information can be roughly divided into two types, static traffic information and dynamic traffic information. Dynamic traffic information collection technology is divided into automatic and non-automatic collection according to its collection method. The main feature of non-automatic collection technology is that it requires manual intervention to complete the collection of traffic information, such as manual collection method, test vehicle mobile survey, etc.; the main feature of automatic collection technology is that it completely relies on the collection equipment to automatically sense the passage or existence of road users, realize the all-round and real-time collection of traffic information, and has more application value, so the method that can automatically collect traffic information in real time is the main discussion direction.

Traditional methods of collecting traffic information include on-site observation by traffic investigators, manual statistics, traffic counters and traffic cameras. However, they have limitations such as low data collection efficiency and poor real-time performance. Traffic information collection technology mostly uses a single sensor with a single data type and many limitations. At present, according to the different principles of information collection equipment, the types of equipment used on the roadside mainly include magnetic frequency sensors, wave frequency sensors, piezoelectric sensors, video sensors and GPS.

In contrast, modern technology has brought about tremendous changes in traffic information collection. Global Positioning System (GPS) technology can track the location, speed and direction of vehicles in real time through devices installed on vehicles, providing highly accurate and real-time data. The widespread use of smartphones and mobile applications has enabled individuals to become participants in information collection, anonymously providing data such as vehicle speed, driving trajectory and traffic congestion, although attention needs to be paid to privacy and data quality issues. In addition, the deployment of sensors and high-resolution cameras supported by the Internet of Things technology combined with computer vision technology has made automated traffic information collection and processing possible, including vehicle detection, license plate recognition and traffic violation monitoring. These modern technologies provide powerful tools to improve the efficiency, accuracy and real-time nature of traffic information collection, helping to better manage traffic flow, monitor congestion and ensure traffic safety.

At present, common public databases for vehicle trajectory data in ramp merging areas are mostly built based on cameras, and most of them are recorded by drones, such as the NGSIM and INTERACTION datasets. These databases usually collect data from the vehicle-mounted perspective as the main perspective, such as the Mirror-Traffic and Lyft datasets. There is a lack of public datasets of vehicle trajectories that focus on ramp merging areas from the roadside perspective.

Summary of the invention

In order to address the deficiencies of the prior art, the present invention provides a method and system for generating a vehicle trajectory database in a ramp merging area, which selectively utilizes the roadside point cloud data of the ramp merging area and realizes the construction of a comprehensive and accurate vehicle trajectory database through vehicle target detection and trajectory tracking processing.

In order to achieve the above object, the present invention adopts the following technical solution:

In a first aspect, the present invention provides a method for generating a vehicle trajectory database in a ramp merging area.

A method for generating a vehicle trajectory database in a ramp merging area includes the following steps:

Obtain LiDAR point cloud data of ramp merging areas;

Identify all vehicle targets based on LiDAR point cloud data;

The trajectories of all identified vehicle targets are tracked respectively, the driving data of each vehicle target is calculated according to the trajectory tracking results of each vehicle target, and the vehicle trajectory database is generated according to the trajectory tracking results and driving data of each vehicle target.

As a further limitation of the first aspect of the present invention, identifying all vehicle targets according to the laser radar point cloud data includes:

The laser radar point cloud data is segmented to obtain the three-dimensional bounding box of the vehicle target, the point cloud features in the three-dimensional bounding box are pooled, and the pooled point cloud features are regularized to obtain each vehicle target.

As a further limitation of the first aspect of the present invention, segmenting the laser radar point cloud data to obtain a three-dimensional bounding box of the vehicle target includes:

According to the X-axis and Z-axis of the LiDAR coordinate system, the LiDAR data is divided into multiple discrete bins, and a search range is set for each bin. Each one-dimensional search distance is equally divided into multiple bins to characterize the center position of different targets. The three-dimensional bounding box is regressed using the cross entropy loss function to obtain the final three-dimensional bounding box of the vehicle target.

As a further limitation of the first aspect of the present invention, performing pooling processing on the point cloud features within the three-dimensional bounding box includes:

Define the parameters of each 3D bounding box ,in, Represent the center coordinates of the three-dimensional bounding box, Represents the height, width and length of the three-dimensional bounding box, Indicates the orientation angle of the 3D bounding box;

After expanding the three-dimensional bounding box, the parameters of the expanded three-dimensional bounding box are obtained. , is the preset constant value;

The true value of the segmentation mask obtained by point cloud segmentation is judged The point cloud in the image belongs to the foreground point or the background point, and the 3D bounding box that does not contain the foreground point is eliminated.

As a further limitation of the first aspect of the present invention, the point cloud features after the pooling process are subjected to a canonical transformation, and in the transformed canonical coordinate system, the origin of the coordinates is located at the center of the three-dimensional bounding box cluster, and the canonical coordinate system and The axis is parallel to the ground and The direction is consistent with the vehicle's forward direction. The axis is consistent with the Y axis of the laser radar coordinate system;

If the parameters of the 3D bounding box are , the parameters of the true 3D bounding box are , after regular transformation:

, ,in, Represent the center coordinates of the real 3D bounding box, Represents the height, width and length of the real 3D bounding box, Indicates the orientation angle of the real 3D bounding box;

Loss Function for:

,in, represents all generated 3D bounding boxes, Represents the candidate clusters that are considered as positive samples in the 3D bounding box regression calculation, is the confidence level of the target estimate, is the true value of the estimated target, is the cross information entropy loss, and Represent bin classification loss and residual regression loss respectively.

As a further limitation of the first aspect of the present invention, the AB3DMOT algorithm is used to track the trajectories of all identified vehicle targets respectively.

In a second aspect, the present invention provides a system for generating a vehicle trajectory database in a ramp merging area.

A system for generating a vehicle trajectory database in a ramp merging area, comprising:

The data acquisition unit is configured to: acquire laser radar point cloud data of a ramp merging area;

The vehicle target recognition unit is configured to: recognize all vehicle targets according to the laser radar point cloud data;

The database generation unit is configured to: perform trajectory tracking on all identified vehicle targets respectively, calculate the driving data of each vehicle target according to the trajectory tracking results of each vehicle target, and generate a vehicle trajectory database according to the trajectory tracking results and driving data of each vehicle target.

In a third aspect, the present invention provides a ramp merging area vehicle trajectory database, which is generated by the ramp merging area vehicle trajectory database generation method described in the first aspect of the present invention; it can be a ramp merging area vehicle trajectory database for a set time period, or data can be imported in real time to form a ramp merging area vehicle trajectory database for a long period of time.

In a fourth aspect, the present invention provides a computer device, comprising: a processor and a computer-readable storage medium;

a processor adapted to execute a computer program;

A computer-readable storage medium, wherein a computer program is stored in the computer-readable storage medium, and when the computer program is executed by the processor, the method for generating a ramp merging area vehicle trajectory database as described in the first aspect of the present invention is implemented.

In a fifth aspect, the present invention provides a computer-readable storage medium storing a computer program, wherein the computer program is suitable for being loaded by a processor and executing the method for generating a ramp merging area vehicle trajectory database as described in the first aspect of the present invention.

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

1. The present invention innovatively proposes a method and system for generating a vehicle trajectory database in a ramp merging area, which selects and utilizes the roadside point cloud data of the ramp merging area to track the trajectories of all identified vehicle targets respectively, calculates the driving data of each vehicle target according to the trajectory tracking results of each vehicle target, and generates a vehicle trajectory database according to the trajectory tracking results and driving data of each vehicle target, thereby realizing the construction of a comprehensive and accurate vehicle trajectory database.

2. The present invention segments the laser radar point cloud data to obtain a three-dimensional bounding box of the vehicle target, performs pooling processing on the point cloud features within the three-dimensional bounding box, performs regularization transformation on the pooled point cloud features to obtain each vehicle target, thereby ensuring the accuracy of vehicle target recognition.

3. The present invention divides the surrounding area into multiple discrete bins according to the X-axis and Z-axis, sets a search range for each bin, divides each one-dimensional search distance into multiple bins to characterize the center position of different targets, and uses the cross entropy loss function to perform regression calculation on the three-dimensional bounding box to ensure the accuracy of three-dimensional bounding box recognition.

4. The present invention introduces the distance from each point to the coordinate origin into the model as an additional feature. Such distance information helps to restore or compensate for the loss of depth information in the subsequent network learning process, ensuring that the model can accurately understand the three-dimensional structure of the target object.

5. The loss function of the present invention combines classification loss and regression loss to achieve accurate optimization of the three-dimensional bounding box, which not only improves the accuracy of the model, but also enhances the model's ability to locate and estimate the size of objects in complex scenes.

Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention.

BRIEF DESCRIPTION OF THE DRAWINGS

The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

FIG1 is a schematic diagram of a direct acceleration lane merging area provided in Embodiment 1 of the present invention;

FIG2 is a schematic diagram of a parallel acceleration lane merging area provided in Embodiment 1 of the present invention;

FIG3 is a schematic diagram of a data acquisition platform for a laser radar provided in Example 1 of the present invention;

FIG. 4 is a diagram showing the visualization effect of vehicle trajectories provided in Example 1 of the present invention.

DETAILED DESCRIPTION

The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

Embodiment 1:

This implementation proposes a method for generating a vehicle trajectory database in a ramp merging area, including the following process:

S1: Build a data collection platform based on LiDAR to collect traffic data of ramp merging areas and save it in LiDAR point cloud data format;

S2: Process the LiDAR point cloud data and identify all vehicle targets;

S3: Track the trajectories of all identified vehicle targets, calculate the data of other fields (such as speed and acceleration) based on the trajectory tracking results, and establish a comprehensive vehicle trajectory database.

In step S1 of this implementation, the definition domain classification of the ramp merging area includes:

Vehicle merging refers to the process in which two traffic flows coming from the same direction merge into one. On the highway, the merging area consists of three parts: the entrance ramp, the acceleration lane and the main lane. The entrance ramp is used to guide vehicles into the main road; the acceleration lane is located on the right side of the main lane, providing acceleration space for vehicles entering the ramp so that they can safely merge into the main traffic flow; the main lane is separated by a central isolation strip for straight driving.

According to the different designs of the acceleration lanes, the merging area can be divided into direct acceleration lanes (as shown in Figure 1) and parallel acceleration lanes (as shown in Figure 2). Taking the three lanes of the inner lane, the middle lane and the outer lane as an example, the section where the merging area is located is the acceleration lane, and the acceleration lane includes an acceleration section and a gradient section. The direct acceleration lane is mainly suitable for sections with less traffic. It gradually extends diagonally from the ramp and directly connects to the main lane, but the driver's driving comfort may be reduced due to the steep connection. In comparison, the parallel acceleration lane is more suitable for sections with more traffic. The acceleration section is parallel to the main lane, providing a smoother driving trajectory and clear lane line division, making it easier to drive and identify. The present invention selects the merging area of the parallel acceleration lane for research (as shown in Figure 2).

In step S1 of this implementation, a data acquisition platform based on laser radar is built, including:

Since there is usually no directly available AC power supply on the road section where data is collected, it is necessary to carry a portable mobile power supply for outdoor power support. The equipment involved in building a data collection platform based on LiDAR includes: batteries, inverters, host computers (usually laptops), LiDAR and its bracket, power lines, and data lines. The connection method is shown in Figure 3. The parameters of the LiDAR should not be lower than the requirements in Table 1.

Table 1: LiDAR parameters

In step S2 of this implementation, vehicle target recognition includes:

Before tracking the vehicle target, the vehicle must first be accurately detected, which involves processing the original LiDAR 3D point cloud data to identify a single target vehicle. These preliminary detection results are then used to guide the vehicle's trajectory tracking process. To this end, the present invention adopts the PointRCNN algorithm to perform high-precision target detection on the 3D point cloud data generated by the LiDAR. The algorithm is mainly implemented through the following three key steps: first, generating target candidate clusters, then performing pooling processing on the point cloud area, and finally fine-tuning and normalizing the 3D bounding box.

Generate target candidate clusters based on point cloud segmentation, including:

The present invention firstly utilizes the PointNet++ network to perform deep feature extraction on point cloud data with the help of its multi-scale grouping characteristics; then, a point feature learning strategy is adopted to effectively distinguish the foreground points in the scene through the "segmentation head", and then the "box regression head" is used to process the original point cloud data encoded by the backbone network to estimate the foreground mask and generate a three-dimensional candidate box.

In the point cloud segmentation stage, the present invention generates the corresponding segmentation mask true value by utilizing the true value of the three-dimensional bounding box annotation in the data set. Considering the imbalance of the number of foreground and background points in the actual scene, the present invention introduces Focalloss (FL for short), which is expressed as shown in formula (1), where: Represents the predicted probability of the prospect point, parameter Set to 0.25, Set to 2 to optimize the point cloud segmentation effect:

FL( )= (1);

If it is a foreground point, the predicted probability The value of , otherwise, the predicted probability The value of .

In this implementation, a bin-based 3D candidate box generation strategy is preferably designed, which divides the surrounding area into multiple discrete bins according to the X-axis and Z-axis, and sets a search range S for each bin; then, each one-dimensional search distance is equally divided into multiple bins to characterize the center position of different targets; finally, the cross entropy loss function is used to perform regression calculation on the 3D bounding box.

Point cloud area pooling, including:

In the point cloud region pooling stage, the present invention performs a detailed pooling process on the point cloud features within the previously identified and generated three-dimensional bounding box, aiming to further extract the detailed location information of the target object. This process involves the precise definition and adjustment of each three-dimensional bounding box to ensure that the most representative feature information can be extracted from it. The parameters defining each three-dimensional bounding box can be expressed by formula (2):

(2);

in, Represent the center coordinates of the bounding box, Represents the height, width and length of the bounding box, Indicates the orientation angle of the bounding box. In order to capture richer semantic information, the present invention appropriately expands the bounding box, which not only helps to enhance the model's perception of the details of the target object, but also provides more comprehensive data support for subsequent accurate detection. The expanded bounding box is expressed as formula (3):

(3);

in, is a preset constant value used to adjust the size of the bounding box to include more possible foreground points while retaining enough background information for comparison. By analyzing the true value of the segmentation mask previously obtained through point cloud segmentation, the present invention can determine The point cloud in the image belongs to the foreground point or the background point. This step significantly improves the quality of the candidate clusters of bounding boxes. The bounding boxes that do not contain foreground points will be eliminated to ensure that only the areas that may contain the target object are processed later, thereby achieving accurate estimation and recognition of the target object's position.

Standardized 3D bounding box optimization, including:

In view of the above-mentioned 3D bounding box optimization process, the present invention adopts a more accurate canonical transformation method to process the point cloud data obtained by point cloud region pooling. This method aims to transform the point cloud data into a more standardized coordinate system for more accurate bounding box positioning and size estimation. In the transformed canonical coordinate system, the transformed coordinate system has the following characteristics: (1) the origin of the coordinate is located at the center of the candidate cluster of the bounding box, (2) the canonical coordinate system and The axis is parallel to the ground and ensure The direction is consistent with the vehicle's forward direction, (3) The axes are consistent with the LiDAR coordinate system. This coordinate system setting provides a standardized reference frame for subsequent bounding box optimization.

In order to compensate for the loss of depth information that may be caused by coordinate transformation, the present invention introduces the distance from each point to the coordinate origin as an additional feature into the model. Such distance information helps to restore or compensate for the loss of depth information in the subsequent network learning process, ensuring that the model can accurately understand the three-dimensional structure of the target object. This distance information can be expressed as shown in formula (4):

(4)

Furthermore, the present invention performs a canonical coordinate transformation on the 3D bounding box of each object and sets the 3D intersection-over-union (IOU) threshold to 0.5 to optimize the accuracy of the bounding box and ensure the maximum overlap between the candidate cluster of the bounding box and the real bounding box, thereby accurately reflecting the position and size of the object. If the parameters of the candidate cluster of the bounding box are expressed as , and the corresponding true bounding box parameters are , then after regularization transformation, these parameters will be converted into standardized form as shown in formula (5):

(5);

This transformation helps to uniformly handle different bounding box candidate clusters and simplifies the subsequent optimization and training process.

The final loss function combines classification loss and regression loss to optimize the performance of the model, as shown in formula (6):

(6).

Here, represents all generated 3D bounding box candidate clusters, and represents the candidate clusters that are considered as positive samples in the bounding box regression calculation. is the estimated confidence level of the target; is the true value of the estimated target; is the cross information entropy loss; and Can be and The calculations show that, respectively, they represent the bin classification loss and the residual regression loss, which jointly guide the network training process to achieve accurate optimization of the 3D bounding box. The design of this comprehensive loss function not only improves the accuracy of the model, but also enhances the model's ability to locate and estimate the size of objects in complex scenes.

In step S3 of this implementation, the vehicle trajectory tracking method includes:

The present invention uses the AB3DMOT algorithm as the core mechanism for processing point cloud data trajectory tracking. In order to achieve continuous tracking of dynamic targets and maintain data consistency and accuracy, a three-dimensional Kalman filter module is used, combined with the Hungarian algorithm to achieve cross-frame target tracking and data association. The algorithm comprehensively uses five key modules to achieve high-precision tracking and analysis of moving targets, including:

(1) 3D object detection module, responsible for the preliminary identification and positioning of various objects such as vehicles and pedestrians;

In order to achieve accurate detection of three-dimensional target point clouds, the present invention adopts the PointRCNN algorithm, which can identify and locate vehicle targets from point cloud data, providing accurate initial data for subsequent trajectory tracking. The detection results can be organized into a set, namely （ Represents the total number of detected targets), where each detected target is represented by a tuple , in this collection Indicates the center position coordinates of the target, Represents the three-dimensional size of the target, which are length, width and height. Represents the target's direction angle, The confidence score of the target detection is obtained by accurately identifying the spatial position, size and orientation of each target. PointRCNN not only provides a reliable foundation for trajectory tracking, but also lays the foundation for further behavior analysis and scene understanding.

(2) A three-dimensional Kalman filter module that uses historical data to predict the current state of the target to achieve continuous tracking;

Since the Kalman filter algorithm was proposed in 1960, it has been widely used in many fields due to its powerful linear system state estimation capability. Specifically, the core assumptions of this algorithm module include: ① The tracked system follows a linear dynamic model; ② The noise that affects the measurement can be assumed to be white noise and follows a Gaussian distribution. Based on these assumptions, the Kalman filter algorithm formula is now derived and demonstrated to continuously track the state of the target object.

If from Time has come At this moment, it is assumed that the state equation of the prediction system is as follows:

(7);

in, is the current state estimate, is the state transition matrix, is the state at the previous moment, is the control input gain matrix, is the system input vector, The mean is 0 and the covariance matrix is And the process noise follows a normal distribution, indicating the uncertainty in the model prediction.

In summary, the Kalman filter process is an effective two-stage iterative mechanism (prediction and update), which provides a methodological basis for the continuous estimation of the target state in a dynamic system. In each iteration, the Kalman filter first uses the previous state estimate to predict the state of the target at the current moment, and then uses the newly obtained observation data to correct this prediction in order to obtain a more accurate state estimate.

In target tracking applications, especially for dynamic targets in three-dimensional space such as vehicles and pedestrians, the Kalman filter algorithm shows its unique advantages. By setting the state vector including the target's spatial position, size, orientation, speed and other information, the Kalman filter can effectively track the target and accurately predict its motion trajectory in three-dimensional space. The state prediction equation and state update equation in this process are based on the mathematical description of the target motion model, allowing the algorithm to continuously predict and update the length, width, height and other dimensional information of the center coordinate value of the target to be tracked under limited observation errors and process noise, so as to obtain the motion trajectory of the target to be tracked.

In this module, the Kalman filter algorithm is used to accurately track the trajectory of a moving target in three-dimensional space, and a constant velocity model is used to describe the dynamic behavior of the target. By constructing a comprehensive state vector, the present invention can continuously update the estimate of the target's motion state, thereby achieving continuous tracking. In a specific implementation, the target's motion state is expressed as a set of parameters containing information such as its position, orientation, size, and speed. The present invention defines the state vector as: , used to describe the trajectory information of the moving target, where In order to predict the state of the target at a certain moment in the future, the present invention uses a constant velocity model for simplified calculation, which can be expressed as formula (8):

(8);

Here They represent the target position predicted at the next moment. In the trajectory tracking process of each frame, from the previous frame Pass to current frame Track information is the key input of the tracking algorithm. This information is expressed as a set of trajectories ,in Represents the frame The number of trajectories identified in the set The present invention is based on the constant velocity model prediction of the Kalman filter algorithm and calculates its The predicted state is shown in formula (9):

(9);

Here The complete state vector predicted for the i -th target in the current frame, including information such as position, orientation, size, and speed.

(3) Data association module, which performs matching work by analyzing the results of target detection and Kalman filter prediction;

In solving the problem of multi-target tracking, the key is how to accurately identify and track multiple targets in the scene, especially in continuous video frames. This challenge is essentially an association problem, which requires determining which targets are consistent in the previous and subsequent frames. The present invention uses the Hungarian algorithm to solve the problem of multi-target tracking. It constructs a cost matrix to represent the matching cost between each detected target and each predicted trajectory. Then, the algorithm looks for a way to associate the detected targets with the existing trajectories with the minimum total cost.

According to the multi-task assignment model, the mathematical description of the multi-target tracking problem can be shown as (10):

(10);

In the multi-target tracking problem, the moving targets between adjacent frames have a unique matching relationship, and there will not be a target in one frame matching multiple targets in another frame, so the assignment matrix is a 0-1 decision variable, indicating the task Assigned to an employee ; is the cost matrix, which means that the task Assign to staff The cost or price incurred when the target is matched. In the multi-target trajectory tracking problem, the higher the similarity of the targets to be matched, the smaller the cost value. If the similarity is lower, the difference between the two targets is more obvious, and the cost is greater. In general, the cost matrix and the benefit matrix are relative. If we assume represents the benefits after task assignment. The goal at this time is to find an assignment method that minimizes the total cost of all tasks to employees. Based on the above mathematical model, the relevant total cost maximization can be expressed as formula (11):

(11);

Set the minimum cost matrix to:

(12);

Therefore, the benefit matrix is shown in formula (13):

(13);

The optimal solution with the minimum cost can be expressed as:

(14);

because is a constant, then:

(15);

Right now:

(16);

From equations (15) and (16), it can be seen that this process ensures the equivalence between cost minimization and benefit maximization, thereby improving the accuracy and efficiency of multi-target tracking.

In this module, by creating a dimension The present invention can calculate the matching similarity between each pair of detected targets and predicted trajectories, and then use the three-dimensional intersection of Union (3D IoU) to provide a specific score for each pair of detected targets and predicted trajectories. In this way, by solving the Hungarian algorithm, the matched and unmatched targets can be determined, providing a reliable basis for subsequent trajectory updates. The relevant data association results can be expressed as follows using a mathematical description:

(17);

In formula (18), and Represent the successfully matched trajectory and the detected target respectively, reflects the total number of targets that have been successfully associated in the data. and They represent the trajectories and targets for which no matches were found, respectively.

(4) Target object state update module, which adjusts the target tracking state according to the new observation data;

In the target object state update module, the trajectory tracking system optimizes the estimate of the target state by integrating the detection information of each target and the corresponding prediction information. Specifically, the system updates the target trajectory based on the data of frame t. These trajectories are represented by the set Indicates that for each matched trajectory, the updated state is expressed as (in ), which not only reflects the latest position and attributes of the target, but also integrates the detection information and original trajectory information ,In this process, the weighted average method is used, where the weight coefficient is provided by the Kalman filter algorithm, ,to ensure that the update process takes into account the latest observation data and improves the ,continuity and reliability of tracking.

(5) Trajectory lifecycle management module, which automatically manages the generation, maintenance, and termination process of trajectories.

The trajectory lifecycle management module is a crucial part of the trajectory tracking system. It is responsible for handling the entry and exit of the target object in the monitoring scene, ensuring that the system can flexibly adapt to the dynamic changes of the target object. This module mainly includes two key links: ① New target recognition and trajectory initialization: For all detection results that fail to match the existing trajectory , the system regards them as newly appeared targets and marks them. In order to avoid data redundancy caused by occasional false detections or other factors, these unmatched detection results will not immediately generate new trajectory data. Only when these detection results are continuously identified and successfully associated and matched in multiple consecutive frames will the system create new trajectories for them. ②Trajectory termination and data cleaning: For those existing trajectories that fail to find matching objects in the new frame, The system marks them as targets that are about to exit the monitoring scene. Considering that short-term occlusion or detection failure may cause the trajectory to lose matching in a short period of time, the system introduces a trajectory retention cycle , which is the maximum time that a track can be retained before failing to re-match the target. If the track still fails to find a matching target within this time window, the system considers that the target has left the scene and clears the corresponding track data from the system.

In step S3 of this implementation, other fields are calculated and a trajectory database is established, including:

In order to improve the database, the present invention not only tracks the trajectory of the target vehicle, but also further analyzes the speed and acceleration of the vehicle, and manually connects partially disconnected trajectories, thereby improving the continuity and accuracy of the data.

(1) The process of calculating the target vehicle’s speed is as follows:

By using trajectory tracking technology, the present invention can obtain the speed of the target vehicle along the x-axis, y-axis and z-axis in the laser radar coordinate system in each frame. The actual driving speed of the vehicle can be calculated by formula (18):

(18);

here, Indicates the actual speed of the vehicle. , and Represent the speed of the vehicle in the x-axis, y-axis and z-axis directions respectively.

(2) The process of solving the acceleration of the target vehicle is as follows:

Since the same vehicle may appear in different consecutive frames, the present invention can calculate the acceleration of the vehicle by comparing the speed change between two consecutive frames. The calculation process is shown in formula (19):

(19);

in, Represents the acceleration of the vehicle in each frame, is the speed of the next frame, is the speed of the previous frame, assuming the time interval between each frame is 0.1 seconds.

Through in-depth analysis and calculation, the present invention not only successfully tracks the dynamic trajectory of the vehicle, but also accurately obtains the speed and acceleration information of the vehicle, which has important practical application value for in-depth understanding of vehicle driving behavior, optimizing traffic flow, and improving road safety. In addition, by manually connecting the disconnected trajectories, the continuity and integrity of the data are significantly enhanced, providing more accurate and comprehensive data support for subsequent traffic flow research and analysis. Based on the above data, a vehicle trajectory database called Roadside LiDAR Dataset (RSLD) can be established.

This implementation provides the following examples:

An interchange merging area at a certain place was selected for data collection. The total length of the acceleration lane in the merging area is about 180 meters, with a total of four lanes, namely the inner lane, the middle lane, the outer lane and the acceleration lane.

In order to collect relevant information of road users in the merging area, a data collection platform was built. The collection platform mainly includes laser radar, laptop computer, mobile power supply and transformer. The LeiShen LS-C32 laser radar is used. The mobile power supply is a 24V lithium battery, and its voltage is increased to 220V through a transformer to ensure the normal operation of the laser radar and laptop computer.

According to the local traffic operation characteristics, 7:30-9:00 in the morning and 17:30-18:30 in the afternoon on weekdays were selected as peak hours for data collection, and 9:30-10:30 in the morning and 15:00-16:30 in the afternoon were selected as off-peak hours to collect off-peak data in the merging area. A total of 10 hours of traffic flow data were collected. According to the above steps, based on the AB3DMOT trajectory tracking algorithm, the tracking effect is shown in Figure 4, which shows the results of vehicle trajectory tracking in the 115th frame, the 120th frame, the 123rd frame and the 129th frame.

Therefore, a roadside LiDAR dataset (RSLD) was established, which includes ten sub-databases, including five off-peak and five peak-period sub-databases, and a total of 8,830 vehicle trajectory data are recorded, of which 2,233 are off-peak and 6,597 are peak-period trajectory data.

Embodiment 2:

This implementation provides a system for generating a ramp merging area vehicle trajectory database, including:

The data acquisition unit is configured to: acquire the laser radar point cloud data of the ramp merging area; the specific working process of this unit is the same as the detailed process in step S1 in embodiment 1, and will not be repeated here;

The vehicle target recognition unit is configured to: recognize all vehicle targets according to the laser radar point cloud data; the specific working process of this unit is the same as the detailed process in step S2 in embodiment 1, which will not be repeated here;

The database generation unit is configured to: track the trajectories of all identified vehicle targets respectively, calculate the driving data of each vehicle target according to the trajectory tracking results of each vehicle target, and generate a vehicle trajectory database according to the trajectory tracking results and driving data of each vehicle target; the specific working process of this unit is the same as the detailed process in step S3 in Example 1, and will not be repeated here.

It is understandable that the above-mentioned units can be separately or completely combined into one or several other units to constitute, or one (some) of the units can be further divided into multiple smaller units in function to constitute, which can achieve the same operation without affecting the realization of the technical effects of the embodiments of the present application. The above-mentioned units are divided based on logical functions. In practical applications, the functions of one unit can also be implemented by multiple units, or the functions of multiple units can be implemented by one unit. In other embodiments of the present application, the database generation system may also include other units. In practical applications, these functions can also be implemented with the assistance of other units, and can be implemented by the collaboration of multiple units.

According to another embodiment of the present application, the system described in this embodiment can be constructed, and the database generation method of the embodiment of the present application can be implemented by running a computer program (including program code) capable of executing the steps involved in the corresponding method described in Example 1 on a general computing device such as a computer including processing elements and storage elements such as a central processing unit (CPU), random access memory (RAM), and read-only memory (ROM). The computer program can be recorded on, for example, a computer-readable recording medium, and loaded into the above-mentioned computing device through the computer-readable recording medium and run therein.

Embodiment 3:

The present implementation provides a ramp merging area vehicle trajectory database, which is generated by the ramp merging area vehicle trajectory database generation method described in Example 1 of the present invention; it can be a ramp merging area vehicle trajectory database for a set time period, or data can be imported in real time to form a ramp merging area vehicle trajectory database for a long period of time.

Embodiment 4:

This implementation provides an electronic device, which includes a processor, a communication interface, and a computer-readable storage medium, wherein the processor, the communication interface, and the computer-readable storage medium may be connected via a bus or other means.

Among them, the communication interface is used to receive and send data, the computer-readable storage medium can be stored in the memory of the electronic device, the computer-readable storage medium is used to store a computer program, the computer program includes program instructions, and the processor is used to execute the program instructions stored in the computer-readable storage medium.

A processor (or CPU (Central Processing Unit)) is the computing core and control core of an electronic device, which is suitable for implementing one or more instructions, and specifically suitable for loading and executing one or more instructions to implement corresponding method flows or corresponding functions.

The processor is configured to perform the following process:

Obtaining the laser radar point cloud data of the ramp merging area; the specific working process is the same as the detailed process in step S1 in Example 1, which will not be repeated here;

All vehicle targets are identified according to the laser radar point cloud data; the specific working process is the same as the detailed process in step S2 in Example 1, which will not be repeated here;

The trajectories of all identified vehicle targets are tracked respectively, the driving data of each vehicle target is calculated according to the trajectory tracking results of each vehicle target, and a vehicle trajectory database is generated according to the trajectory tracking results and driving data of each vehicle target; the specific working process is the same as the detailed process in step S3 in Example 1, and will not be repeated here.

Embodiment 5:

This embodiment provides a computer-readable storage medium (Memory), which is a memory device in an electronic device for storing programs and data. It is understandable that the computer-readable storage medium here can include both built-in storage media in the electronic device and, of course, extended storage media supported by the electronic device. The computer-readable storage medium provides a storage space that stores the processing system of the electronic device.

In addition, the storage space also stores one or more instructions suitable for being loaded and executed by the processor, and these instructions may be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here may be a high-speed RAM memory, or a non-volatile memory (non-volatile memory), such as at least one disk storage; optionally, it may also be at least one computer-readable storage medium located away from the aforementioned processor.

In one embodiment, the computer-readable storage medium stores one or more instructions; the processor loads and executes the one or more instructions stored in the computer-readable storage medium to implement the following process:

Obtaining the laser radar point cloud data of the ramp merging area; the specific working process is the same as the detailed process in step S1 in Example 1, which will not be repeated here;

All vehicle targets are identified according to the laser radar point cloud data; the specific working process is the same as the detailed process in step S2 in Example 1, which will not be repeated here;

The trajectories of all identified vehicle targets are tracked respectively, the driving data of each vehicle target is calculated according to the trajectory tracking results of each vehicle target, and a vehicle trajectory database is generated according to the trajectory tracking results and driving data of each vehicle target; the specific working process is the same as the detailed process in step S3 in Example 1, and will not be repeated here.

A person skilled in the art can appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. A person skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on the computer, the process or function according to the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted through a computer-readable storage medium. The computer instructions can be transmitted from a website site, a computer, a server or a data center to another website site, a computer, a server or a data center by wired (for example, coaxial cable, optical fiber, digital line (DSL)) or wireless (for example, infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by the computer or a data processing device such as a server, a data center, etc. that includes one or more available media integrated. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a tape), an optical medium (for example, a DVD), or a semiconductor medium (for example, a solid-state drive (SSD)), etc.

The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims (5)

1. A method for generating a ramp merging area vehicle trajectory database, characterized by comprising the following steps: Obtain LiDAR point cloud data of ramp merging areas; Identify all vehicle targets based on LiDAR point cloud data; Track all identified vehicle targets respectively, calculate the driving data of each vehicle target according to the tracking results of each vehicle target, and generate a vehicle trajectory database according to the tracking results and driving data of each vehicle target; All vehicle targets are identified based on the LiDAR point cloud data, including: Segment the laser radar point cloud data to obtain the three-dimensional bounding box of the vehicle target, perform pooling on the point cloud features within the three-dimensional bounding box, and perform regular transformation on the pooled point cloud features to obtain each vehicle target; Segment the LiDAR point cloud data to obtain the three-dimensional bounding box of the vehicle target, including: According to the X-axis and Z-axis of the laser radar coordinate system, the laser radar data is divided into multiple discrete bins, and the search range is set for each bin. Each one-dimensional search distance is equally divided into multiple bins to represent the center position of different targets. The three-dimensional bounding box is regressed using the cross entropy loss function to obtain the final three-dimensional bounding box of the vehicle target. Pooling is performed on the point cloud features within the 3D bounding box, including: Define the parameters of each 3D bounding box ,in, Represent the center coordinates of the three-dimensional bounding box, Represents the height, width and length of the three-dimensional bounding box, Indicates the orientation angle of the 3D bounding box; After expanding the three-dimensional bounding box, the parameters of the expanded three-dimensional bounding box are obtained. , is the preset constant value; The true value of the segmentation mask obtained by point cloud segmentation is judged Whether the point cloud in the foreground point or the background point, the 3D bounding box that does not contain the foreground point is eliminated; The point cloud features after pooling are transformed into regular coordinates. In the transformed regular coordinate system, the origin of the coordinates is located at the center of the three-dimensional bounding box cluster. and The axis is parallel to the ground and The direction is consistent with the vehicle's forward direction. The axis is consistent with the Y axis of the laser radar coordinate system; If the parameters of the 3D bounding box are , the parameters of the true 3D bounding box are , after regular transformation: , ,in, Represent the center coordinates of the real 3D bounding box, Represents the height, width and length of the real 3D bounding box, Indicates the orientation angle of the real 3D bounding box; Loss Function for: ,in, represents all generated 3D bounding boxes, Represents the candidate clusters that are considered as positive samples in the 3D bounding box regression calculation, is the confidence level of the target estimate, is the true value of the estimated target, is the cross information entropy loss, and Represent bin classification loss and residual regression loss respectively; The AB3DMOT algorithm is used to track the trajectories of all identified vehicle targets. 2. A system for generating a vehicle trajectory database for a ramp merging area, implementing the method for generating a vehicle trajectory database for a ramp merging area according to claim 1, characterized in that it comprises: The data acquisition unit is configured to: acquire laser radar point cloud data of a ramp merging area; The vehicle target recognition unit is configured to: recognize all vehicle targets according to the laser radar point cloud data; The database generation unit is configured to: perform trajectory tracking on all identified vehicle targets respectively, calculate the driving data of each vehicle target according to the trajectory tracking results of each vehicle target, and generate a vehicle trajectory database according to the trajectory tracking results and driving data of each vehicle target. 3. A ramp merging area vehicle trajectory database, characterized in that it is generated using the ramp merging area vehicle trajectory database generation method according to claim 1. 4. A computer device, comprising: a processor and a computer readable storage medium; a processor adapted to execute a computer program; A computer-readable storage medium, wherein a computer program is stored in the computer-readable storage medium, and when the computer program is executed by the processor, the method for generating a ramp merging area vehicle trajectory database according to claim 1 is implemented. 5. A computer-readable storage medium, characterized in that the computer-readable storage medium stores a computer program, and the computer program is suitable for being loaded by a processor and executing the method for generating a ramp merging area vehicle trajectory database as claimed in claim 1.