Vehicle obstacle intention prediction method, vehicle trajectory prediction method and device

Abstract

The application discloses a vehicle obstacle intention prediction method, a vehicle trajectory prediction method and equipment. The vehicle obstacle intention prediction method comprises the following steps: obtaining a preset number of target obstacles corresponding to a target vehicle; predicting the lateral intention and the initial longitudinal intention of each target obstacle by using the position information and the state information of each target obstacle; determining the interaction degree between any two target obstacles based on each lateral intention, and obtaining an interaction graph corresponding to all target obstacles based on all the interaction degrees; predicting the target obstacle with a yielding intention based on the interaction graph; determining the target obstacle with both the yielding intention and the initial longitudinal intention, determining the final longitudinal intention of the target obstacle by using the yielding intention; and taking the lateral intention, the yielding intention and the final longitudinal intention as the intention prediction result of the target obstacle. In this way, the intention of the vehicle obstacle can be predicted.

Description

Technical Field

[0001] This application relates to the field of autonomous driving technology, and in particular to methods and devices for predicting vehicle obstacle intentions, vehicle trajectory prediction, and related equipment. Background Technology

[0002] Currently, for autonomous vehicles to navigate safely and efficiently in complex traffic conditions dominated by human drivers, they need the ability to proactively make decisions, such as deciding when to change lanes, overtake, or slow down to allow other vehicles to merge. This requires autonomous vehicles to know the future trajectories of surrounding vehicles, enabling them to plan their own driving in advance and avoid danger. Furthermore, when multiple obstacles exist around an autonomous vehicle, the future behavior of these obstacles may influence each other, thus affecting their future trajectories.

[0003] Therefore, predicting the potential future intentions between obstacles is of great significance. Summary of the Invention

[0004] This application provides a method for predicting vehicle obstacle intent, a method for predicting vehicle trajectory, and an apparatus. By sequentially acquiring the lateral intent, interaction map, yielding intent, and final longitudinal intent of the target obstacle, the intent prediction result of the target obstacle is obtained. The method focuses on the interaction between obstacles, thereby enabling the prediction of the intent of vehicles surrounding the target vehicle.

[0005] To address the aforementioned technical problems, this application provides a method for predicting the intent of vehicle obstacles, comprising: acquiring a preset number of target obstacles corresponding to a target vehicle; predicting the lateral intent and initial longitudinal intent of each target obstacle using its position and state information; determining the interaction degree between any two target obstacles based on each lateral intent, and obtaining an interaction graph corresponding to all target obstacles based on all interaction degrees; predicting target obstacles with yielding intent based on the interaction graph; identifying target obstacles with both yielding intent and initial longitudinal intent, and determining the final longitudinal intent of the target obstacle using the yielding intent; and using the lateral intent, yielding intent, and final longitudinal intent as the intent prediction results of the target obstacles.

[0006] The process of obtaining a preset number of target obstacles corresponding to the target vehicle includes: obtaining several initial obstacles within a preset range of the target vehicle; and selecting a preset number of target obstacles from the several initial obstacles.

[0007] The process of selecting a preset number of target obstacles from a number of initial obstacles includes: determining the obstacle score for each target obstacle by using the distance, orientation, relative longitudinal speed, and relative lateral speed between each target obstacle and the target vehicle, the distance between each target obstacle and adjacent obstacles, and the correlation of obstacle confidence; and filtering all target obstacles according to their obstacle scores to select a preset number of target obstacles.

[0008] The location information includes the lateral position information relative to the lane, and the status information includes historical trajectory features and the heading angle relative to the lane.

[0009] The process involves using the position and status information of each target obstacle to predict its lateral intention, including: determining the deviation between the heading angle and a preset heading angle; if the deviation is less than a deviation threshold, predicting that the target obstacle has no lateral intention; if the deviation is greater than or equal to the deviation threshold, determining whether the lateral position information is within the lane change area corresponding to the heading angle; if so, predicting the lane change probability based on historical trajectory features; if the lane change probability is greater than a threshold, predicting that the target obstacle has a lateral intention.

[0010] Specifically, based on each lateral intent, the interaction degree between any two target obstacles is determined, including: determining the relative state of any two target obstacles using the lateral intents corresponding to the two target obstacles; determining the numerical value corresponding to the relative state, and using the numerical value as the interaction degree.

[0011] Specifically, based on all interaction degrees, an interaction graph corresponding to all target obstacles is obtained, including: taking the target obstacles as interaction nodes in the interaction graph, and taking the interaction degree between any two target obstacles as the weight between the corresponding interaction nodes; the node value of an interaction node is the sum of the interaction degrees between the interaction node and the other interaction nodes.

[0012] The process of predicting target obstacles with yielding intentions based on the interaction graph includes: traversing each interaction node in the interaction graph and determining the interaction domain based on the interaction degree between interaction nodes; performing a decision search on all interaction nodes contained in the interaction domain and predicting target obstacles with yielding intentions.

[0013] The process of traversing each interaction node in the interaction graph and determining the interaction domain based on the interaction degree between interaction nodes includes: traversing the interaction nodes from high to low according to their node values, and adding interaction nodes with an interaction degree greater than a threshold to the interaction domain.

[0014] The process involves performing a decision search on all interactive nodes within the interaction domain to predict target obstacles with yielding intentions. This includes: traversing all interactive nodes in the interaction domain from highest to lowest node value, performing a decision search on the target interactive node for each traversal, and obtaining the corresponding decision variables; traversed target interactive nodes do not participate in subsequent decision searches; and using the decision variables to determine the target obstacles with yielding intentions.

[0015] The initial longitudinal intention includes the intention to maintain the original speed, the intention to accelerate, or the intention to decelerate.

[0016] The determination of the final longitudinal intention of the target obstacle using the yielding intention includes: if at least one of the intention to maintain the original speed, the intention to accelerate, and the intention to decelerate satisfies the constraint of the yielding intention, then the intention to maintain the original speed, the intention to accelerate, and the intention to decelerate with the highest priority is selected as the final longitudinal intention of the target obstacle; if the intention to maintain the original speed, the intention to accelerate, and the intention to decelerate cannot satisfy the constraint of the yielding intention, then the final longitudinal intention of the target obstacle is the intention to maintain the original speed.

[0017] To solve the above-mentioned technical problems, another technical solution adopted in this application is to provide a vehicle trajectory prediction method, which includes: using the above-mentioned vehicle obstacle intent prediction method to obtain the intent prediction result of the vehicle obstacle corresponding to the target vehicle, and then predicting the vehicle trajectory of the target vehicle based on the intent prediction result.

[0018] To solve the above-mentioned technical problems, another technical solution adopted in this application is to provide an electronic device, which includes a memory and a processor, wherein the memory is used to store a computer program, and the processor is used to execute the computer program to implement the above-mentioned vehicle obstacle intention prediction method or vehicle trajectory prediction method.

[0019] To solve the above-mentioned technical problems, another technical solution adopted in this application is to provide a computer-readable storage medium for storing a computer program, which, when executed by a processor, is used to implement the above-mentioned vehicle obstacle intention prediction method or vehicle trajectory prediction method.

[0020] The beneficial effects of this application are as follows: Unlike existing technologies, the vehicle obstacle intent prediction method provided in this application obtains a preset number of target obstacles corresponding to the target vehicle; uses the position and state information of each target obstacle to predict the lateral intent and initial longitudinal intent of each target obstacle; based on each lateral intent, determines the interaction degree between any two target obstacles, and based on all interaction degrees, obtains the interaction graph corresponding to all target obstacles; based on the interaction graph, predicts target obstacles with yielding intent; identifies target obstacles that simultaneously have yielding intent and initial longitudinal intent, and uses the yielding intent to determine the final longitudinal intent of the target obstacle; and uses the lateral intent, yielding intent, and final longitudinal intent as the intent prediction results of the target obstacle. Through the above method, the lateral intent, yielding intent, and final longitudinal intent of the target obstacle can be considered more comprehensively, and the yielding intent is obtained based on the lateral intent, and the final longitudinal intent is obtained based on the yielding intent, making the correlation between the lateral intent, yielding intent, and final longitudinal intent stronger, thus improving the accuracy of obstacle intent prediction. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0022] Figure 1 This is a flowchart illustrating the first embodiment of the vehicle obstacle intention prediction method provided in this application;

[0023] Figure 2 The orientation division provided in this application Figure 1 Schematic diagram of the embodiment;

[0024] Figure 3 This is a flowchart illustrating an embodiment of step 12 provided in this application;

[0025] Figure 4 This is a schematic diagram of an embodiment of the possible lane change area corresponding to the relative heading angle provided in this application;

[0026] Figure 5 This application provides the interaction Figure 1 Schematic diagram of the embodiment;

[0027] Figure 6 This is a schematic diagram of an embodiment of the interaction domain provided in this application;

[0028] Figure 7 This is a flowchart illustrating an embodiment of step 14 provided in this application;

[0029] Figure 8 This is a flowchart illustrating an embodiment of the vehicle trajectory prediction method provided in this application;

[0030] Figure 9 This is a schematic diagram of the structure of an embodiment of the electronic device provided in this application;

[0031] Figure 10 This is a schematic diagram of an embodiment of the computer-readable storage medium provided in this application. Detailed Implementation

[0032] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0033] See Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the vehicle obstacle intent prediction method provided in this application. The method includes:

[0034] Step 11: Obtain the preset number of target obstacles corresponding to the target vehicle.

[0035] In some embodiments, a number of initial obstacles within a preset range of the target vehicle are acquired, and then a preset number of target obstacles are selected from the initial obstacles. The obstacles are vehicles other than the target vehicle.

[0036] Specifically, the distance, orientation, relative longitudinal speed, and relative lateral speed between each target obstacle and the target vehicle, the distance between each target obstacle and adjacent obstacles, and the correlation of obstacle confidence can be used to determine the obstacle score corresponding to each target obstacle. Then, all target obstacles can be filtered according to their obstacle scores to select a preset number of target obstacles.

[0037] The formulas used to determine the obstacle score for each obstacle include:

[0038]

[0039]

[0040]

[0041]

[0042]

[0043] Among them, S i The score is obtained by weighted summation of six data points: distance to the target vehicle, location, longitudinal velocity relative to the target vehicle, lateral velocity relative to the target vehicle, average distance to the six nearest target obstacles, and perceived information confidence level; w1, w2, w3, w4, and w5 are weights; i represents the i-th target obstacle to be scored; d max,ego This represents the maximum distance between the target obstacle and the target vehicle; d i,ego This represents the distance between the i-th target obstacle and the target vehicle; v s,i,ref v represents the longitudinal velocity of the i-th target obstacle relative to the target vehicle; d,i,ref d represents the lateral velocity of the i-th target obstacle relative to the target vehicle; i,j d represents the distance between the i-th target obstacle and its nearest j-th target obstacle; max Indicates the preset d i,j Maximum value; con i θi represents the confidence level of the i-th target obstacle; θ1, θ2, and θ3 represent preset weights, which can be adjusted according to actual applications; l min,1 l min,2 and l min,3 This indicates the preset upper / lower function threshold, which can be adjusted according to the actual application.

[0044] Using the above formula, we can obtain the obstacle score for each target obstacle. Then, we can sort them in descending order to select a preset number of target obstacles.

[0045] In some embodiments, based on the lane and position of the target vehicle, the lane where the target vehicle is located and the two lanes adjacent to the lane where the target vehicle is located are divided into six directions, such as... Figure 2 As shown, with the target vehicle as the center, six directions are determined: front, rear, left front, left rear, right front, and right rear. Then, the search extends along these six directions until a preset distance threshold (e.g., 50 meters) is exceeded. During this extension, if a lane branch is encountered, a parallel search begins from the new branch. After all searches are completed, the lanes traversed during the extension are marked. Finally, all obstacles around the target vehicle are traversed. If the target vehicle's lane is within a marked lane, it is selected. When all obstacles in marked lanes are selected, and the number of selected obstacles has not reached the preset obstacle filtering limit, other obstacles are sorted, and the highest-ranked obstacles are added to the valid target filtering sequence until all obstacles are processed, or the number of selected obstacles reaches the preset limit (e.g., 70 obstacles).

[0046] Step 12: Using the position and status information of each target obstacle, predict the lateral intent and initial longitudinal intent of each target obstacle.

[0047] Understandably, predicting the intent of a target obstacle requires predicting lateral intent. Lateral intent indicates whether the target obstacle intends to change lanes. This is because when a vehicle changes lanes, it needs to move laterally to the left or right to complete the lane change.

[0048] In some embodiments, corresponding feature values ​​can be extracted based on the state information and lane information of each target obstacle, so as to predict the lateral intent based on the feature values.

[0049] The longitudinal intent is used to indicate whether the target obstacle continues to travel in the current direction.

[0050] In some embodiments, the location information includes the lateral position information of the target obstacle relative to the lane, where the lateral position information may be the lateral position of the target obstacle relative to the left and right boundary lines and the center line of the lane. The status information includes the historical trajectory characteristics of the target obstacle and the heading angle of the target obstacle relative to the lane. See also... Figure 3 Step 12 may include the following process:

[0051] Step 31: Determine the deviation between the heading angle and the preset heading angle.

[0052] In some embodiments, the presence of a lane change intention is determined by comparing the difference between the heading angle of the target obstacle relative to the lane and a preset heading angle with a deviation threshold.

[0053] Step 32: Determine whether the deviation is less than the deviation threshold.

[0054] like Figure 4 As shown, Figure 4 This is a schematic diagram of possible lane change areas corresponding to the relative heading angle. The shaded area is the possible lane change area, and the three lines are the left and right boundary lines and the middle line of the lane where the obstacle is located.

[0055] In some embodiments, the relationship between the deviation of the heading angle of the relative lane and a preset heading angle and a deviation threshold is first determined by judgment. If the deviation of the heading angle and the preset heading angle is less than the deviation threshold, step 33 is executed; if the deviation of the heading angle and the preset heading angle is greater than or equal to the deviation threshold, step 34 is executed. The deviation threshold can be a threshold between angles, such as 0° to 10°.

[0056] Step 33: Predict that the target obstacle does not have lateral intent.

[0057] Specifically, if the deviation between the heading angle and the preset heading angle is less than the deviation threshold, it is considered that the target obstacle has no intention to change lanes, that is, the target obstacle is predicted to have no lateral intention.

[0058] In some embodiments, the presence of lateral intent can be indicated by corresponding identifiers. For example, 1 indicates the presence of lateral intent, and 0 indicates the absence of lateral intent.

[0059] Step 34: Determine whether the lateral position information is within the lane change area corresponding to the heading angle.

[0060] Specifically, if the deviation between the heading angle and the preset heading angle is greater than or equal to the deviation threshold, it is necessary to further determine whether the lateral position of the target obstacle is within the possible lane-changing area corresponding to the relative heading angle (e.g., Figure 4 If the shaded area is present, then proceed to step 35.

[0061] Step 35: Predict the lane change probability based on the historical trajectory features. If the lane change probability is greater than the threshold, the target obstacle is predicted to have lateral intent.

[0062] In some embodiments, when the lateral position of the target obstacle is within the possible lane change area corresponding to the relative heading angle, it is necessary to input historical trajectory features (such as the average value of speed, acceleration, heading angle and relative lane line distance corresponding to the most recent 5 to 10 trajectory points) into the MLP (Multilayer Perceptron) to calculate the lane change probability of the target obstacle based on the MLP. When the lane change probability is greater than a preset value (such as 80% or 90%), it is considered that the target obstacle has the intention to change lanes, and at this time the lateral intention of the target obstacle can be obtained.

[0063] Step 13: Based on each lateral intent, determine the interaction degree between any two target obstacles, and based on all interaction degrees, obtain the interaction graph corresponding to all target obstacles.

[0064] In some embodiments, the relative state of any two target obstacles is determined by the lateral intent corresponding to any two target obstacles, and then the numerical value corresponding to the relative state is determined, and the numerical value is used as the interaction degree.

[0065] In some embodiments, the interactivity of each selected target obstacle can be calculated pairwise based on the acquired lateral intent. The calculation is performed by determining whether the relative state of the target obstacle belongs to a preset state set, and assigning different values ​​accordingly. The preset state set can be S = [{Car A changes lanes, Car B stays, and the target lane for the lane change is ahead of the target lane for staying}, {Car A changes lanes, Car B changes lanes, and the target lane for the lane change is the other vehicle's original lane}, {Car A changes lanes, Car B changes lanes, and the target lane for the lane change is the same lane}, {Car A stays, Car B stays, and they are in the same lane, and the vehicle behind is moving faster}] or others. The preset state set can be expanded or deleted according to needs or actual circumstances. Furthermore, corresponding values ​​can be set for elements in the state set. When a relative state matches any element in the state set, the value of that element is obtained as the value corresponding to the relative state.

[0066] In some embodiments, target obstacles are treated as interaction nodes in the interaction graph, and the interaction degree between any two target obstacles is used as the weight between the corresponding interaction nodes, wherein the node value of an interaction node is the sum of the interaction degrees between the interaction node and the other interaction nodes.

[0067] like Figure 5 As shown, each obstacle is treated as a node, the interaction degree between nodes (between target obstacles) is used as the edge weight, and the sum of the interaction degrees between a node and all other nodes is the node value. This process is repeated to construct the interaction graph. Figure 5 In the example, there are 5 obstacles A, B, C, D, and E. The interaction degree corresponding to these obstacles is determined using the method described above, and an interaction graph is constructed using these interaction degrees. For example... Figure 5 As shown, the node values ​​of the five obstacles (interaction nodes) A, B, C, D, and E are 2, 1, 1, 1, and 1, respectively. There is interaction between A and C, between A and B, and between D and E. In other words, A and C, A and B, and D and E are "connected". Figure 6 As shown, the interaction domain only contains A, B, C, and E.

[0068] Step 14: Based on the interaction graph, predict the target obstacle with the intention to give way.

[0069] In some embodiments, a higher degree of interaction between obstacles indicates that the obstacles are more likely to exhibit similar behaviors in the future, such as changing lanes towards the same lane. Therefore, target obstacles with yielding intentions can be further predicted based on the obstacle's location and status information.

[0070] In some embodiments, see Figure 7 Step 14 may include the following process:

[0071] Step 71: Traverse each interaction node in the interaction graph and determine the interaction domain based on the degree of interaction between the interaction nodes.

[0072] In some embodiments, the interaction nodes can be traversed from high to low according to their node values, and interaction nodes with an interaction degree greater than a threshold can be added to the interaction domain.

[0073] Specifically, the interaction nodes in the interaction graph are sorted from high to low according to their node values ​​and traversed. Each time the graph is traversed, an interaction domain is established with that interaction node as the center.

[0074] The rule for establishing the interaction domain is to start from the initial interaction node, consider an interaction degree greater than 0 as "connected", and an interaction degree equal to 0 as "not connected". This rule is used to search for connected components, and the connected component nodes obtained by the search constitute the interaction domain.

[0075] Step 72: Perform a decision search on all interaction nodes in the interaction domain to predict the target obstacle with the intention to yield.

[0076] In some embodiments, all interactive nodes in the interactive domain are traversed from high to low according to their node values. A decision search is performed on the target interactive node for each traversal to obtain the corresponding decision variables. The target interactive nodes that have already been traversed do not participate in subsequent decision searches. Then, the decision variables are used to determine the target obstacle with the intention to yield.

[0077] Specifically, a decision search is performed on all interaction nodes in the interaction domain. The decision search method involves traversing the interaction nodes in the domain from highest to lowest value. Each traversal requires solving an optimization problem to find the optimal combination of decision variables. It is important to note that previously solved target obstacles (interaction nodes) are not included in subsequent traversals, achieving a pruned search effect.

[0078] Among them, the optimization problems that need to be solved in decision search involve the following formulas:

[0079]

[0080]

[0081]

[0082]

[0083]

[0084]

[0085]

[0086] Where k represents the number of target obstacles to be interactively searched for; L represents the time penalty; P1 is the preemption margin penalty; P2 represents the yielding cost penalty; x represents the decision variable, representing the choice of yielding intention of the current target obstacle relative to the i-th target obstacle to be interacted with, x∈{-1,1}, x=1 indicates yielding, x=-1 indicates preemption; d front d represents the longitudinal distance between two target obstacles involved in the interaction; safe Indicates a safe distance; d hori This represents the lateral distance between two target obstacles involved in the interaction; v d,0 This represents the initial lateral velocity of the current target obstacle; v i,s This represents the initial longitudinal velocity of the obstacle to be interacted with; v s,0 The initial longitudinal velocity of the current target obstacle is represented by θ1 and θ2; θ1 and θ2 represent preset weights; k1 represents the weight coefficient, which can be adjusted according to actual application; T max Indicates the preset maximum feasible lane change time; T min This indicates the preset shortest feasible lane change time.

[0087] Step 15: Identify the target obstacle that simultaneously has a yielding intention and an initial longitudinal intention, and use the yielding intention to determine the final longitudinal intention of the target obstacle.

[0088] In some embodiments, the initial longitudinal intention includes an intention to maintain the original speed, an intention to accelerate, or an intention to decelerate. If at least one of the intentions to maintain the original speed, accelerate, and decelerate satisfies the constraint of the yielding intention, then the intention to maintain the original speed, accelerate, and decelerate with the highest priority is selected as the final longitudinal intention of the target obstacle; if none of the intentions to maintain the original speed, accelerate, and decelerate satisfy the constraint of the yielding intention, then the final longitudinal intention of the target obstacle is the intention to maintain the original speed.

[0089] Specifically, by traversing all target obstacles, each traversal requires comparing the yielding intentions of target obstacles to determine whether there is an intersection between them. That is, three intentions (maintain original speed, accelerate, and decelerate) are selected sequentially from the initial longitudinal intentions to check whether the intersection constraint is satisfied. If multiple intentions satisfy the constraint, the intention with the highest priority among maintaining original speed, accelerating, and decelerating is selected as the final longitudinal intention of the target obstacle, with the priority being "maintain original speed > decelerate > accelerate". If none of the three intentions satisfy the constraint, the longitudinal intention is taken as maintaining original speed, and the information is recorded and handed over to downstream processing.

[0090] The aforementioned constraints can be determined using the specific intent of the yielding intention. For example, the yielding intention includes yielding and overtaking. When the yielding intention is to yield, the longitudinal intention of the target obstacle should be to maintain its original speed or decelerate; when the yielding intention is to overtake, the longitudinal intention of the target obstacle should be to accelerate. Therefore, these conditions can be used as constraints to determine the final longitudinal intention of the target obstacle.

[0091] Step 16: Use the lateral intent, yielding intent, and final longitudinal intent as the intent prediction results for the target obstacle.

[0092] Understandably, predicting the intention of a target vehicle to an obstacle requires predicting not only the lateral intention, but also the yielding intention and the final longitudinal intention, so that the lateral intention, yielding intention, and final longitudinal intention can be used as the intention prediction results of the target obstacle.

[0093] Unlike existing technologies, the vehicle obstacle intent prediction method provided in this application predicts the lateral intent and initial longitudinal intent of each target obstacle by utilizing its position and state information. Then, based on each lateral intent, the interaction degree between any two target obstacles is determined, and an interaction graph corresponding to all target obstacles is obtained based on all interaction degrees. Based on the interaction graph, target obstacles with yield intent are predicted. Target obstacles with both yield intent and initial longitudinal intent are identified, and the final longitudinal intent of the target obstacle is determined using the yield intent. By using the lateral intent, yield intent, and final longitudinal intent as the intent prediction results for the target obstacle, this method can more comprehensively consider the lateral intent, yield intent, and final longitudinal intent of the target obstacle. Furthermore, the yield intent is derived from the lateral intent, and the final longitudinal intent is derived from the yield intent, resulting in a stronger correlation between the lateral intent, yield intent, and final longitudinal intent, thus improving the accuracy of obstacle intent prediction.

[0094] See Figure 8 , Figure 8 This is a flowchart illustrating an embodiment of the vehicle trajectory prediction method provided in this application. The method includes:

[0095] Step 81: Obtain the intent prediction results of the vehicle obstacles corresponding to the target vehicle.

[0096] In some embodiments, the intention prediction result of the vehicle obstacle corresponding to the target vehicle can be obtained by the vehicle obstacle intention prediction method of any of the above embodiments.

[0097] The intention prediction results for obstacles include lateral intention, yielding intention, and longitudinal intention.

[0098] Step 82: Predict the vehicle trajectory of the target vehicle based on the intent prediction results.

[0099] It is understandable that the trajectory of a target vehicle can be predicted based on whether the obstacle indicates an intention to change lanes, give way, accelerate, decelerate, or maintain its original speed.

[0100] Unlike existing technologies, the vehicle trajectory prediction method provided in this application can predict the trajectory of the target vehicle by obtaining the intention prediction results of the vehicle obstacles corresponding to the target vehicle. Since the intention prediction results have high accuracy, the accuracy of the target vehicle trajectory prediction is improved accordingly.

[0101] See Figure 9 , Figure 9 This is a schematic diagram of an embodiment of the electronic device provided in this application. The electronic device 90 includes a memory 901 and a processor 902. The memory 901 is used to store computer programs, and the processor 902 is used to execute the computer programs to implement the vehicle obstacle intention prediction method or vehicle trajectory prediction method of any of the above embodiments, which will not be described in detail here.

[0102] Among them, electronic device 90 can be an in-vehicle electronic device that can be applied to the field of autonomous driving technology.

[0103] See Figure 10 , Figure 10 This is a schematic diagram of an embodiment of the computer-readable storage medium provided in this application. The computer-readable storage medium 100 is used to store a computer program 1001. When the computer program 1001 is executed by a processor, it is used to implement the vehicle obstacle intention prediction method or the vehicle trajectory prediction method of any of the above embodiments, which will not be described in detail here.

[0104] In summary, the vehicle obstacle intent prediction method provided in this application can more comprehensively consider the lateral intent, yielding intent, and final longitudinal intent of the target obstacle. Furthermore, the yielding intent is derived from the lateral intent, and the final longitudinal intent is derived from the yielding intent, resulting in a stronger correlation between the lateral intent, yielding intent, and final longitudinal intent, thus improving the accuracy of obstacle intent prediction. The vehicle trajectory prediction method provided in this application predicts the trajectory of the target vehicle by obtaining the intent prediction results of the vehicle obstacles corresponding to the target vehicle. Because the intent prediction results have high accuracy, the accuracy of the target vehicle trajectory prediction is correspondingly improved, such as when the target vehicle performs automatic acceleration, deceleration, or avoidance operations during autonomous driving.

[0105] The processor involved in this application may be referred to as a CPU (Central Processing Unit), which may be an integrated circuit chip, or a general-purpose processor, digital signal processor (DSP), application-specific integrated circuit (ASIC), field-programmable gate array (FPGA), or other programmable logic device, discrete gate or transistor logic device, or discrete hardware component.

[0106] The storage media used in this application include various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), or optical discs.

[0107] The above description is merely an embodiment of this application and does not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A method for predicting the intent of a vehicle obstacle, characterized in that, The method includes: Obtain a preset number of target obstacles corresponding to the target vehicle; Using the position and state information of each target obstacle, predict the lateral intent and initial longitudinal intent of each target obstacle; Based on each of the lateral intentions, the interaction degree between any two of the target obstacles is determined, and based on all the interaction degrees, the interaction graphs corresponding to all the target obstacles are obtained. Traverse each interaction node in the interaction graph and determine the interaction domain based on the degree of interaction between the interaction nodes; A decision search is performed on all interactive nodes contained in the interactive domain to predict target obstacles with yielding intentions; Identify a target obstacle that simultaneously possesses the yielding intention and the initial longitudinal intention, and use the yielding intention to determine the final longitudinal intention of the target obstacle; The lateral intent, the yielding intent, and the final longitudinal intent are used as the intent prediction results for the target obstacle.

2. The method according to claim 1, characterized in that, The acquisition of a preset number of target obstacles corresponding to the target vehicle includes: Acquire several initial obstacles within a preset range of the target vehicle; A predetermined number of target obstacles are selected from the initial obstacles.

3. The method according to claim 2, characterized in that, The step of selecting a preset number of target obstacles from the plurality of initial obstacles includes: By utilizing the distance, orientation, relative longitudinal velocity, and relative lateral velocity between each target obstacle and the target vehicle, as well as the distance between each target obstacle and adjacent obstacles and the correlation of obstacle confidence, the obstacle score corresponding to each target obstacle is determined. All the target obstacles are filtered according to their scores, and a preset number of target obstacles are selected.

4. The method according to claim 1, characterized in that, The location information includes lateral position information relative to the lane, and the status information includes historical trajectory features and heading angle relative to the lane; The step of predicting the lateral intent of each target obstacle using its position and state information includes: Determine the deviation between the heading angle and the preset heading angle; If the deviation is less than the deviation threshold, it is predicted that the target obstacle does not have lateral intent. If the deviation is greater than or equal to the deviation threshold, then it is determined whether the lateral position information is within the lane change area corresponding to the heading angle; If so, perform lane change probability prediction on the historical trajectory features. If the lane change probability is greater than a threshold, then predict that the target obstacle has lateral intention.

5. The method according to claim 1, characterized in that, Determining the interaction degree between any two target obstacles based on each lateral intent includes: The relative state of any two target obstacles is determined by utilizing the lateral intent corresponding to any two target obstacles; Determine the numerical value corresponding to the relative state, and use the numerical value as the interaction degree.

6. The method according to claim 5, characterized in that, The step of obtaining the interaction graph corresponding to all the target obstacles based on all the interaction degrees includes: The target obstacle is used as an interaction node in the interaction graph, and the interaction degree between any two target obstacles is used as the weight between the corresponding interaction nodes; the node value of the interaction node is the sum of the interaction degrees between the interaction node and the other interaction nodes.

7. The method according to claim 1, characterized in that, The step of traversing each interaction node in the interaction graph and determining the interaction domain based on the interaction degree between interaction nodes includes: The interaction nodes are traversed from high to low according to their node values, and the interaction nodes with an interaction degree greater than the threshold are added to the interaction domain.

8. The method according to claim 1, characterized in that, The step of performing a decision search on all interaction nodes contained in the interaction domain to predict target obstacles with yielding intentions includes: For all interactive nodes contained in the interactive domain, the nodes are traversed from high to low according to their node values. A decision search is performed on the target interactive node for each traversal to obtain the corresponding decision variables. The target interactive nodes that have already been traversed do not participate in the subsequent decision search. The decision variables are used to identify target obstacles that have the intention to give way.

9. The method according to claim 1, characterized in that, The initial longitudinal intent includes an intent to maintain the original speed, an intent to accelerate, or an intent to decelerate. Determining the final longitudinal intent of the target obstacle using the yielding intent includes: If at least one of the intention to maintain original speed, the intention to accelerate, and the intention to decelerate satisfies the constraint of the intention to give way, then the intention to maintain original speed, the intention to accelerate, and the intention to decelerate with the highest priority is selected as the final longitudinal intention of the target obstacle. If the intention to maintain the original speed, the intention to accelerate, and the intention to decelerate all fail to satisfy the constraint of the intention to give way, then the final longitudinal intention of the target obstacle is to maintain the original speed.

10. A vehicle trajectory prediction method, characterized in that, The method includes: The intention prediction result of vehicle obstacles corresponding to the target vehicle is obtained by using the method as described in any one of claims 1-9; The trajectory of the target vehicle is predicted based on the intent prediction result.

11. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory being used to store a computer program and the processor being used to execute the computer program to implement the method as described in any one of claims 1-10.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store a computer program, which, when executed by a processor, is used to implement the method as described in any one of claims 1-10.

Images