Road participant trajectory prediction method considering environmental information
By combining the parallelism detection and kinematic model prediction of road information and target historical trajectory, long-term trajectory prediction of long-distance targets in complex traffic environments is achieved, and the problems of low accuracy and high calculation costs in the prior art are solved. Improves prediction accuracy and accuracy.
Patent Information
- Application Number
- CN202410978248.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-22
- Publication Date
- 2024-12-27
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The prior art cannot effectively predict the long-term trajectory of long-distance road participants in complex dynamic traffic environments, especially the situation of incoming vehicles, resulting in low accuracy of long-distance target trajectory prediction in low-computing power schemes and calculation based on deep learning methods. High cost and insufficient real-time.
By combining road information and the historical trajectory of the target, the parallelism calculation method is used to detect whether the target follows the road, and predict short-term and long-term trajectories with kinematic models, the weighted average method is used to integrate the prediction results to improve prediction accuracy.
It improves the accuracy and accuracy of low-cost road target trajectory prediction, can effectively make decisions on the maneuvering behavior of road participants, and meets the long-term trajectory prediction needs in complex traffic environments.
Smart Images
Figure BDA0004955230310000031 
Figure BDA0004955230310000034 
Figure BDA0004955230310000035
Abstract
Description
Technical Field
[0001] The invention relates to a method for predicting the trajectory of road participants taking environmental information into consideration, and belongs to the technical field of intelligent driving. Background Art
[0002] In order to drive safely in a complex dynamic traffic environment, intelligent driving or advanced driving assistance systems will always calculate the collision probability between the vehicle and other road participants in the future. Therefore, it is necessary to predict the future state of the vehicle and other road participants respectively. For the trajectory prediction of road participants, the kinematic model is currently mainly used to predict the trajectory of the future period based on the current motion state of the target. The kinematic model usually adopts the combination of uniform acceleration and fixed heading angle, uniform acceleration and fixed yaw rate, etc. However, the prediction method based on the inertial motion of the target is similar to the actual driving trajectory of the target only in a short time. Some intelligent driving functions need to consider the collision between the vehicle and the distant target, especially for the oncoming vehicle. Therefore, the long-term trajectory prediction for the distant target is essential, and the trajectory prediction method based only on the kinematic model cannot meet the long-term trajectory requirements. Popular methods based on classical machine learning, deep learning and reinforcement learning can solve the above problems to a certain extent. Most of these methods are strategy-based methods. By extracting the motion characteristics of the target and considering the interaction factors and map information, accurate predictions can be made within a longer prediction range, and they can adapt to more complex environments and longer time ranges. This method can naturally generate multimodal trajectories, which is consistent with the diversity of vehicle maneuvers. However, deep learning-based methods require a large amount of data for training, and the computational cost and time increase dramatically with the increase in the number of factors considered and the number of network layers. Compared with methods based on kinematic models, this type of method can consider more factors, has relatively high accuracy, and has a longer prediction length, but the computational cost is also high, which is insufficient for real-time performance for advanced driver assistance systems that are limited by computing resources. Summary of the invention
[0003] The main purpose of the present invention is to detect the relationship between road participants and the actual road through a parallelism calculation method. For participants who meet the requirement of following the road, a combination of road information and current motion status is used to predict the trajectory over a long period of time, thereby solving the problem of low accuracy in trajectory prediction of long-distance targets in low computing power solutions.
[0004] The purpose of the present invention can be achieved by adopting the following technical solutions:
[0005] A method for predicting the trajectory of road participants taking into account environmental information comprises the following steps:
[0006] Step 1: Combine the trajectory information of the road where the vehicle is located and the historical trajectory of the target to select the effective area, which is defined as D:Droad ∩D object ;
[0007] Step 2: Use the angle difference between the road trajectory and the target trajectory The weight information w(k) along with the speed is accumulated to measure the parallelism of the two curves and determine whether the target maneuver behavior is lane following;
[0008] Step 3: Predict the target's speed in the short term based on the vehicle kinematic model and yaw rate And perform coordinate transformation to obtain the predicted trajectory information in the same coordinate system;
[0009] Step 4: Use the short-term predicted speed information and the trajectory information of the target road.
[0010] Calculate the long-term predicted position x k ,y k , that is, adding time information to the road trajectory;
[0011] Step 5: Use the weighted average method to fuse the results of short-term prediction and long-term prediction.
[0012] Furthermore, the effective area of the road model is the polynomial segment in front of the ego vehicle, defined as D raod :(0,∞), and the effective trajectory of the target is the rear segment of the target, defined as D obj :(obj_pos x ,∞), where obj_pos x is the longitudinal distance of the target in the vehicle trajectory coordinate system. The final valid segment is the intersection of the road segment and the target segment, defined as D:D road ∩D object ;
[0013] The parallelism measure of two trajectories is the angular deviation of the two trajectories. Both trajectories are represented by polylines. The calculation formula is shown in Formula 1.
[0014]
[0015] in:
[0016] n: the number of points on the polyline;
[0017] The heading angle of the road trajectory at point k;
[0018] The heading angle of the target historical trajectory at point k;
[0019] w(k): weight factor for different points.
[0020] Furthermore, the weight information along with the speed is set for more accurate target maneuver detection, as shown in Formula 2.
[0021]
[0022] in:
[0023] c: weight factor
[0024] r(k): radius at point k;
[0025] v: vehicle speed.
[0026] Furthermore, by calculating the road trajectory polynomial at x = obj pso_x The deviation between the heading angle at and the target heading angle is used to detect the maneuverability of the target, as shown in Formula 3.
[0027]
[0028] Among them: roadmodel is the expression of the road trajectory, expressed by a cubic polynomial.
[0029] Furthermore, let the current target motion state be: the vehicle speed Tangential acceleration Yaw rate The steering wheel angle is δ, and the uniform acceleration model is used for longitudinal speed prediction, that is, the target speed predicted in the kth step is:
[0030]
[0031] For lateral motion, since the wheelbase of the target cannot be identified, the target curvature is directly used for calculation, that is, the heading angle predicted in the kth step is:
[0032]
[0033] in:
[0034] Crvt obj : target curvature;
[0035] Assume that the velocity and yaw rate change between two consecutive prediction steps is small, such as Figure 3 As shown, the vector can be calculated by using the average of the corresponding velocity and yaw rate of the two prediction steps as a constant. Size and direction:
[0036]
[0037] Starting vector is located in the target vehicle coordinate system, and Located in the new coordinate system, coordinate system conversion is required for subsequent calculations.
[0038] Going a step further, similar to the short-term forecast definition To long-term predict the motion vector at step k, the target's motion distance is calculated using the speed of the two prediction steps:
[0039]
[0040] Use the road trajectory to calculate the target's direction of motion:
[0041]
[0042] The position of the new trajectory point, which contains not only the position information but also the time information:
[0043]
[0044] y k =f(x k ) (111
[0045] After obtaining the short-term prediction and long-term prediction trajectories, the final target prediction trajectory is obtained through weighted fusion:
[0046]
[0047] in:
[0048] is the kth predicted point of the fusion trajectory;
[0049] is the kth predicted point of the short-term trajectory;
[0050] is the kth predicted point of the long-term trajectory;
[0051] w k is the weight over time, relying on short-term forecasts when time is small and long-term forecasts when time is large.
[0052] Beneficial technical effects of the present invention:
[0053] 1. First, the historical trajectory of the road target is used to check the parallelism with the road environment information to determine whether the target follows the road. Then, the long-term trajectory of the target is predicted by combining the kinematic model and the environment information to meet the goal of following the road. The test results of a large number of road data show that this method improves the accuracy of low-cost road target trajectory prediction.
[0054] 2. By measuring the parallelism between road participants and road trajectories, the maneuvering behavior of road participants is determined; to meet the goal of following the road, the trajectory predicted by the kinematic model and the trajectory predicted by road information are integrated to improve the accuracy of the predicted trajectory of road participants. BRIEF DESCRIPTION OF THE DRAWINGS
[0055] Figure 1 A schematic diagram of screening effective areas of target tracks and road tracks of the present invention;
[0056] Figure 2 A schematic diagram of target maneuver detection based on heading angle of the present invention;
[0057] Figure 3 It is a schematic diagram of trajectory prediction based on kinematic model of the present invention;
[0058] Figure 4 A schematic diagram of a trajectory prediction structure based on road information of the present invention;
[0059] Figure 5 It is a schematic diagram of the trajectory fusion weight coefficient of the present invention;
[0060] Figure 6 This is a strategy flow chart for predicting the trajectory of road participants according to the present invention. DETAILED DESCRIPTION
[0061] In order to make the technical solution of the present invention more clear and specific to those skilled in the art, the present invention is further described in detail below in conjunction with embodiments and drawings, but the implementation manner of the present invention is not limited thereto.
[0062] In a complex dynamic traffic environment, the maneuvering behaviors of road participants are diverse, including: targets traveling in the same direction, targets traveling in the opposite direction, targets crossing, targets turning, etc. Therefore, it is necessary to detect the maneuvering behaviors of targets before using road information to predict the trajectories of road participants. Parallelism calculation is a calculation method used to screen trajectories that can be fused in the lane fusion algorithm. By accumulating the parallelism of each trajectory with other trajectories on the road, the spatial similarity of the two curves is verified in the polar coordinate system. By calculating the parallelism of the historical trajectory of the road participant and the road information, it is possible to determine whether the road participant is following the road.
[0063] like Figure 1 As shown in Figure 2, the effective area of the road model and the target trajectory needs to be screened before the parallelism calculation. The effective area of the road model is the polynomial segment in front of the vehicle, defined as D raod :(0,∞), and the effective trajectory of the target is the rear segment of the target, defined as D obj :(obj_pos x ,∞), where obj_posx is the longitudinal distance of the target in the vehicle trajectory coordinate system. The final valid segment is the intersection of the road segment and the target segment, defined as D:D road ∩D object ;
[0064] The parallelism measure of two trajectories is the angular deviation of the two trajectories. Both trajectories are represented by polylines. The calculation formula is shown in Formula 1.
[0065]
[0066] in:
[0067] n: the number of points on the polyline;
[0068] The heading angle of the road trajectory at point k;
[0069] The heading angle of the target historical trajectory at point k;
[0070] w(k): weight factor of different points;
[0071] Since the trajectories of the nearer segments are more concerned when the ego vehicle is traveling at a low speed, and the trajectories of the farther segments need to be concerned when the ego vehicle is traveling at a high speed, the weight information with speed is set to more accurately detect the target maneuver, as shown in Formula 2.
[0072]
[0073] in:
[0074] c: weight factor
[0075] r(k): radius at point k;
[0076] v: vehicle speed;
[0077] Although the target's historical trajectory is more robust for target trajectory prediction, there are invalid moments in the target's historical trajectory. At this time, the target heading angle is used as an alternative, such as Figure 2 As shown. By calculating the road trajectory polynomial at x = obj pso_x The deviation between the heading angle at and the target heading angle is used to detect the maneuverability of the target, as shown in Formula 3.
[0078]
[0079] Among them: roadmodel is the expression of the road trajectory, expressed by a cubic polynomial;
[0080] After completing the target maneuver detection, trajectory prediction is performed for the target that meets the lane following requirement. Trajectory prediction mainly includes short-term prediction based on kinematic model, long-term prediction based on environmental information, and final trajectory fusion.
[0081] Short-term trajectory prediction is based on uniform acceleration and constant yaw rate. The speed and heading angle of each step are predicted according to the current motion state of the target within a specific prediction time, and then the position of each step in the future in the vehicle coordinate system is obtained through coordinate conversion.
[0082] Assume the current target motion state is: the vehicle speed Tangential acceleration Yaw rate The steering wheel angle is δ, and the uniform acceleration model is used for longitudinal speed prediction, that is, the target speed predicted in the kth step is:
[0083]
[0084] For lateral motion, since the wheelbase of the target cannot be identified, the target curvature is directly used for calculation, that is, the heading angle predicted in the kth step is:
[0085]
[0086] in:
[0087] Crvt obj : target curvature;
[0088] Assume that the velocity and yaw rate change between two consecutive prediction steps is small, such as Figure 3 As shown, the vector can be calculated by using the average of the corresponding velocity and yaw rate of the two prediction steps as a constant. Size and direction:
[0089]
[0090] Starting vector is located in the target vehicle coordinate system, and Located in the new coordinate system, coordinate system conversion is required for subsequent calculations.
[0091] Since the target follows the lane, the road information of the target can be obtained as the lane information of the vehicle translated by a certain lateral distance, and the lateral distance is the lateral deviation of the trajectory of the lane corresponding to the target. Thus, a polynomial y=f(x) can be obtained to represent the lane information of the target, but the trajectory does not contain time information. In order to increase the time information, the speed and time compensation obtained by short-term prediction are combined, and the relevant information can be obtained through an iterative method, such as Figure 4 shown.
[0092] Similar to short-term forecast definition To long-term predict the motion vector at step k, the target's motion distance is calculated using the speed of the two prediction steps:
[0093]
[0094] Use the road trajectory to calculate the target's direction of motion:
[0095]
[0096] Then the position of the new trajectory point contains not only the position information but also the time information:
[0097]
[0098] y k =f(x k ) (11)
[0099] After obtaining the short-term prediction and long-term prediction trajectories, the final target prediction trajectory is obtained through weighted fusion:
[0100]
[0101] in:
[0102] is the kth predicted point of the fusion trajectory;
[0103] is the k-th predicted point of the short-term trajectory;
[0104] is the kth predicted point of the long-term trajectory;
[0105] w k is the weight over time, relying on short-term predictions when time is small and long-term predictions when time is large, such as Figure 5 As shown;
[0106] The logical flow chart of the road participant trajectory prediction can be found in Figure 6 , the specific steps are as follows:
[0107] Step 1: Combine the trajectory information of the road where the vehicle is located and the historical trajectory of the target to select the effective area, which is defined as D:D road ∩D object ;
[0108] Step 2: Use the angle difference between the road trajectory and the target trajectory The weight information w(k) along with the speed is accumulated to measure the parallelism of the two curves and determine whether the target maneuver behavior is lane following;
[0109] Step 3: Predict the target's speed in the short term based on the vehicle kinematic model and yaw rate And perform coordinate transformation to obtain the predicted trajectory information in the same coordinate system;
[0110] Step 4: Use the short-term predicted speed information and the trajectory information of the target road.
[0111] Calculate the long-term predicted position x k ,y k , that is, adding time information to the road trajectory;
[0112] Step 5: Use the weighted average method to fuse the results of short-term prediction and long-term prediction. The weight coefficient is as follows: Figure 5 shown.
[0113] In summary, in this embodiment, according to the method for predicting the trajectory of road participants that takes environmental information into consideration, firstly, by using the historical trajectory of the road target and the road environmental information to perform a parallelism check, it is determined whether the target follows the road. Then, in order to meet the goal of following the road, the long-term trajectory of the target is predicted by combining the kinematic model and environmental information. Tests on a large amount of road data have shown that this method improves the accuracy of low-cost prediction of the trajectory of road targets; by measuring the parallelism between the road participant and the road trajectory, the maneuvering behavior of the road participant is determined. In order to meet the goal of following the road, the trajectory predicted by the kinematic model and the trajectory predicted by the road information are integrated to improve the accuracy of the predicted trajectory of the road participant.
[0114] The above description is only a further embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical solutions and concepts of the present invention within the scope disclosed by the present invention, which belong to the protection scope of the present invention.
Claims
1. A method for predicting the trajectory of road participants taking into account environmental information, characterized in that: The steps include: Step 1: Combine the trajectory information of the road where the vehicle is located and the historical trajectory of the target to select the effective area, which is defined as D: D road ∩D object ; Step 2: Use the angle difference between the road trajectory and the target trajectory The weight information w(k) along with the speed is accumulated to measure the parallelism of the two curves and determine whether the target maneuver behavior is lane following; Step 3: Predict the target's speed in the short term based on the vehicle kinematic model and yaw rate And perform coordinate transformation to obtain the predicted trajectory information in the same coordinate system; Step 4: Use the short-term predicted speed information and the trajectory information of the target road to calculate the long-term predicted position x k ,y k , that is, adding time information to the road trajectory; Step 5: Use the weighted average method to fuse the results of short-term prediction and long-term prediction.
2. A method for predicting the trajectory of road participants taking into account environmental information according to claim 1, characterized in that: The effective area of the road model is the polynomial segment in front of the vehicle, defined as D raod :(0,∞), and the effective trajectory of the target is the rear segment of the target, defined as D obj :(obj_pos x ,∞), where obj_pos x is the longitudinal distance of the target in the vehicle trajectory coordinate system. The final valid segment is the intersection of the road segment and the target segment, defined as D:D road ∩D object ; The parallelism measure of two trajectories is the angular deviation of the two trajectories. Both trajectories are represented by polylines. The calculation formula is shown in Formula 1. in: n: the number of points on the polyline; The heading angle of the road trajectory at point k; The heading angle of the target historical trajectory at point k; w(k): weight factor for different points.
3. The method for predicting the trajectory of road participants considering environmental information according to claim 1, characterized in that: The speed-dependent weight information is set to more accurately detect the target maneuver, as shown in Formula 2. in: c: weight factor r(k): radius at point k; v: vehicle speed.
4. The method for predicting the trajectory of road participants considering environmental information according to claim 1, characterized in that: By calculating the road trajectory polynomial at x = obj pso_x The deviation between the heading angle at and the target heading angle is used to detect the maneuverability of the target, as shown in Formula 3. Among them: roadmodel is the expression of the road trajectory, expressed by a cubic polynomial.
5. The method for predicting the trajectory of road participants considering environmental information according to claim 1, characterized in that: Assume the current target motion state is: the vehicle speed Tangential acceleration Yaw rate The steering wheel angle is δ, and the uniform acceleration model is used for longitudinal speed prediction, that is, the target speed predicted in the kth step is: For lateral motion, since the wheelbase of the target cannot be identified, the target curvature is directly used for calculation, that is, the heading angle predicted in the kth step is: in: Crvt obj : target curvature; Assuming that the speed and yaw rate change between two consecutive prediction steps is small, as shown in Figure 3, the average of the speed and yaw rate corresponding to the two prediction steps can be used as a constant to calculate the vector Size and direction: Starting vector is located in the target vehicle coordinate system, and Located in the new coordinate system, coordinate system conversion is required for subsequent calculations.
6. The method for predicting the trajectory of road participants considering environmental information according to claim 1, characterized in that: Similar to short-term forecast definition To long-term predict the motion vector at step k, the target's motion distance is calculated using the speed of the two prediction steps: Use the road trajectory to calculate the target's direction of motion: The position of the new trajectory point, which contains not only the position information but also the time information: y k =f(x k ) (11) After obtaining the short-term prediction and long-term prediction trajectories, the final target prediction trajectory is obtained through weighted fusion: in: is the kth predicted point of the fusion trajectory; is the kth predicted point of the short-term trajectory; is the kth predicted point of the long-term trajectory; w k is the weight over time, relying on short-term forecasts when time is small and long-term forecasts when time is large.