Driving intention prediction method, device and vehicle

Abstract

The embodiments of the present application disclose a driving intention prediction method, device, and vehicle. The method includes: obtaining a target obstacle for the vehicle; constructing a velocity potential field corresponding to the target obstacle based on the distance between each remaining obstacle on the current road and the target obstacle; obtaining a transformed path for the target obstacle based on the velocity potential field and the predicted driving trajectory of the target obstacle; obtaining multiple differential entropies of the target obstacle based on the transformed path and multiple reference intention paths corresponding to the target obstacle; and obtaining the driving intention of the target obstacle based on a pre-trained hidden Markov model and the multiple differential entropies. The above method can improve the accuracy of the predicted driving intention of the target obstacle.

Description

Technical Field

[0001] The present application relates to the technical field of driving intention prediction, and more specifically, to a driving intention prediction method, device, and vehicle. Background Art

[0002] With the continuous development of autonomous driving technology, its application scenarios are gradually increasing. For example, at intersections of multiple roads, algorithms can be used to predict the driving intentions of surrounding vehicles, allowing the vehicle to plan its own path based on these predicted intentions, thereby improving driving safety. However, these approaches still suffer from inaccurate driving intention prediction. Summary of the Invention

[0003] In view of the above problems, the present application proposes a driving intention prediction method, device and vehicle to improve the above problems.

[0004] In a first aspect, the present application provides a driving intention prediction method, the method comprising: obtaining a target obstacle of a vehicle; constructing a velocity potential field corresponding to the target obstacle based on the distance between each of the remaining obstacles in the current road and the target obstacle, the velocity potential field representing the influence of the driving speed of the remaining obstacles on the driving route of the target obstacle; obtaining a transformation path of the target obstacle based on the velocity potential field and the predicted driving trajectory of the target obstacle; obtaining multiple differential entropies of the target obstacle based on the transformation path and multiple reference intention paths corresponding to the target obstacle, the differential entropies representing the similarity between the transformation path and the corresponding reference intention paths; obtaining the driving intention of the target obstacle based on a pre-trained hidden Markov model and the multiple differential entropies, the driving intention being going straight, turning left, or turning right.

[0005] In a second aspect, the present application provides a driving intention prediction device, which includes: a target obstacle acquisition unit for acquiring a target obstacle of a vehicle; a transformation path acquisition unit for constructing a velocity potential field corresponding to the target obstacle based on the distance between each of the remaining obstacles in the current road and the target obstacle, wherein the velocity potential field represents the influence of the driving speed of the remaining obstacles on the driving route of the target obstacle; obtaining a transformation path of the target obstacle based on the velocity potential field and the predicted driving trajectory of the target obstacle; a differential entropy acquisition unit for obtaining multiple differential entropies of the target obstacle based on the transformation path and multiple reference intention paths corresponding to the target obstacle, wherein the differential entropy represents the similarity between the transformation path and the corresponding reference intention path; a driving intention acquisition unit for obtaining the driving intention of the target obstacle based on a pre-trained hidden Markov model and the multiple differential entropies, wherein the driving intention is to go straight, turn left, or turn right.

[0006] In a third aspect, the present application provides a vehicle comprising one or more processors and a memory; one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs are configured to execute the above-mentioned method.

[0007] In a fourth aspect, the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores program code, wherein the above method is executed when the program code is run.

[0008] The present application provides a driving intention prediction method, device, vehicle, and storage medium. After obtaining a target obstacle of a vehicle, a velocity potential field corresponding to the target obstacle is constructed based on the distance between each remaining obstacle on the current road and the target obstacle, representing the influence of the driving speed of the remaining obstacles on the driving route of the target obstacle. Based on the velocity potential field and the predicted driving trajectory of the target obstacle, a transformed path of the target obstacle is obtained. Based on the transformed path and multiple reference intention paths corresponding to the target obstacle, multiple differential entropies of the target obstacle are obtained, representing the similarity between the transformed path and the corresponding reference intention paths. Based on a pre-trained hidden Markov model and the multiple differential entropies, the driving intention of the target obstacle is obtained, where the driving intention is to go straight, turn left, or turn right. Through the above method, a velocity potential field corresponding to the target obstacle can be constructed based on the distance between each of the remaining obstacles and the target obstacle, and the transformation path of the target obstacle can be obtained based on the velocity potential field and the predicted driving trajectory of the target obstacle. This can take into account the influence of the remaining obstacles on the driving route of the target obstacle, thereby improving the accuracy of the differential entropy obtained based on the transformation path and the reference intention path to characterize the similarity between the two. This makes it possible to obtain the driving intention of the target obstacle based on the accurate differential entropy and the pre-trained hidden Markov model, thereby improving the accuracy of the predicted driving intention of the target obstacle. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0010] Figure 1 A flow chart of a driving intention prediction method proposed in an embodiment of the present application is shown;

[0011] Figure 2 Shown Figure 1A flowchart of an implementation method proposed in S120;

[0012] Figure 3 shows a schematic diagram of a velocity potential field proposed in this application;

[0013] Figure 4 Shown Figure 1 A flowchart of an implementation method proposed in S130;

[0014] Figure 5 A schematic diagram of a transformation path proposed in this application is shown;

[0015] Figure 6 A schematic diagram of a road connection relationship matrix proposed in this application is shown;

[0016] Figure 7 Shown Figure 1 A flowchart of an implementation method proposed in S140;

[0017] Figure 8 A schematic diagram showing a target obstacle driving intention proposed in this application is shown;

[0018] Figure 9 A flowchart of a driving intention prediction method proposed in another embodiment of the present application is shown;

[0019] Figure 10 A structural block diagram of a driving intention prediction device proposed in an embodiment of the present application is shown;

[0020] Figure 11 Shown is a structural block diagram of a vehicle proposed in this application. DETAILED DESCRIPTION

[0021] The following will be combined with the accompanying drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0022] In an embodiment of the present application, the inventors propose a driving intention prediction method, device, and vehicle. After obtaining the target obstacle of the vehicle, based on the distance between each of the remaining obstacles in the current road and the target obstacle, a speed potential field corresponding to the target obstacle is constructed, which characterizes the influence of the driving speed of the remaining obstacles on the driving route of the target obstacle. Based on the speed potential field and the predicted driving trajectory of the target obstacle, the transformation path of the target obstacle is obtained; based on the transformation path and multiple reference intention paths corresponding to the target obstacle, multiple differential entropies of the target obstacle are obtained, which characterize the similarity between the transformation path and the corresponding reference intention paths. Based on a pre-trained hidden Markov model and the multiple differential entropies, the driving intention of the target obstacle is obtained, and the driving intention is to go straight, turn left, or turn right. Through the above method, a velocity potential field corresponding to the target obstacle can be constructed based on the distance between each of the remaining obstacles and the target obstacle, and the transformation path of the target obstacle can be obtained based on the velocity potential field and the predicted driving trajectory of the target obstacle. This can take into account the influence of the remaining obstacles on the driving route of the target obstacle, thereby improving the accuracy of the differential entropy obtained based on the transformation path and the reference intention path to characterize the similarity between the two. This makes it possible to obtain the driving intention of the target obstacle based on the accurate differential entropy and the pre-trained hidden Markov model, thereby improving the accuracy of the predicted driving intention of the target obstacle.

[0023] See also Figure 1 , an embodiment of the present application provides a driving intention prediction method, the method comprising:

[0024] S110: Acquire a target obstacle of the vehicle.

[0025] Among them, the target obstacle of the vehicle may refer to an obstacle on the current road corresponding to the vehicle implementing the driving intention proposed in this application (hereinafter referred to as the own vehicle) for which the driving intention is to be determined.

[0026] As a method, the vehicle can collect environmental information on the current road (such as lanes, pedestrians, vehicles, road signs, etc.) through information collection equipment (such as lidar, cameras, millimeter-wave radar, etc.), and then identify obstacles (such as vehicles around the vehicle) from the collected environmental information based on relevant algorithms (such as target detection algorithms, etc.), and then determine the target obstacle from the identified obstacles.

[0027] In the embodiments of the present application, there may be multiple ways to determine the target obstacle from the identified obstacles.

[0028] As a method of determining the target obstacle, the target obstacle can be determined based on the distance between the identified obstacle and the ego vehicle. Alternatively, the obstacle closest to the ego vehicle can be used as the target obstacle.

[0029] As another way to determine the target obstacle, the target obstacle can be determined based on the identified obstacle's direction of travel and the vehicle's predicted trajectory. Optionally, an obstacle whose trajectory determined based on the direction of travel overlaps with the predicted trajectory can be used as the target obstacle.

[0030] Optionally, the predicted driving trajectory of the ego vehicle or the obstacle may be predicted based on the historical driving trajectory of the ego vehicle or the obstacle.

[0031] S120: Constructing a velocity potential field corresponding to the target obstacle based on the distances between each of the remaining obstacles on the current road and the target obstacle, wherein the velocity potential field represents the influence of the travel speeds of the remaining obstacles on the travel route of the target obstacle.

[0032] The remaining obstacles may refer to all obstacles identified in step S110 except the target obstacle.

[0033] The velocity resultant potential field corresponding to the target obstacle can be understood as a velocity resultant potential field formed based on the velocity potential fields of the remaining obstacles corresponding to the target obstacle.

[0034] As a way, such as Figure 2 As shown, the method of constructing a velocity potential field corresponding to the target obstacle based on the distance between each of the remaining obstacles in the current road and the target obstacle includes:

[0035] S121: Based on the yaw angle, size, speed of each remaining obstacle and the position information of the target obstacle, obtain the distance between each remaining obstacle and the target obstacle.

[0036] The yaw angle can refer to the posture of the remaining obstacle; the size can refer to the length and width of the remaining obstacle's outer rectangle, which can be understood as a rectangle that just encloses the remaining obstacle. The speed can refer to the speed and direction of the remaining obstacle. The position information of the target obstacle can refer to the position of the target obstacle relative to the remaining obstacles, and can be the position information in a coordinate system with the center of the remaining obstacle as the coordinate origin. The center of the remaining obstacle can be understood as the intersection of the diagonals in the outer rectangle.

[0037] As a method, the yaw angle, size, speed of each remaining obstacle and the position information of the target obstacle can be substituted into the relevant calculation formula to obtain the distance between each remaining obstacle and the target obstacle. The relevant calculation formula can be:

[0038]

[0039] Among them, (x0i ,y 0i ) can represent the position information of the target obstacle relative to the i-th (i is a positive integer) remaining obstacle, θ i It can represent the yaw angle of the i-th remaining obstacle, (a i ,b i ) can represent the size of the i-th remaining obstacle, a i It can represent the length of the i-th remaining obstacle, b i It can represent the width of the i-th remaining obstacle, v i It can represent the speed of the i-th remaining obstacle, ζ can represent the velocity potential coefficient, and the velocity potential coefficient can be understood as the potential field morphology correction parameter introduced according to the driving direction of the remaining obstacle.

[0040] Optionally, the reference value of the velocity potential coefficient may be 1.0, and the velocity potential coefficient may be obtained by adjusting the reference value based on the actual situation. Figure 3 As shown in the figure, when the ego vehicle's predicted trajectory overlaps with the predicted trajectory corresponding to a remaining obstacle's direction of travel, that is, the ego vehicle is traveling in the direction of the remaining obstacle, the impact of the remaining obstacle's position on the ego vehicle's trajectory over the next period of time needs to be considered, and the velocity potential coefficient can be 1.0. If the ego vehicle is traveling in the opposite direction of the remaining obstacle's direction of travel, the impact of the remaining obstacle's position on the ego vehicle's trajectory over the next period of time does not need to be considered, and the velocity potential coefficient can be 0. Therefore, the velocity potential field is asymmetric between the direction of travel and the opposite direction of travel, and the velocity potential coefficient can be used to describe this asymmetry of the velocity potential field.

[0041] S122: Obtaining a velocity potential field of each remaining obstacle based on a distance between each remaining obstacle and the target obstacle and a repulsive force range of each remaining obstacle.

[0042] The repulsive force range can be understood as the range formed by the maximum distance that can affect the change of the target obstacle's trajectory when the remaining obstacles continue to move at the current speed.

[0043] Optionally, the repulsive force range of the remaining obstacle can be obtained based on the size of the remaining obstacle and / or the scene type. The larger the size of the remaining obstacle, the larger the repulsive force range of the remaining obstacle.

[0044] As a way, the velocity potential field can be calculated as:

[0045]

[0046] Among them, η can represent the repulsive potential coefficient, which can be understood as a correction parameter for the repulsive range of the remaining obstacles. 0i It can represent the distance between the target obstacle and the i-th (i is a positive integer) remaining obstacle, ρ i It can represent the repulsive force range of the i-th remaining obstacle.

[0047] Optionally, the base value of the repulsive potential coefficient may be 1.0, and the repulsive potential coefficient may be obtained by adjusting the base value based on actual conditions.

[0048] S123: Obtain the velocity potential field based on the velocity potential field of each remaining obstacle.

[0049] As a method, the velocity potential field of each remaining obstacle can be added together to obtain the resulting velocity potential field.

[0050] S130: Obtaining a transformed path of the target obstacle based on the velocity resultant potential field and the predicted driving trajectory of the target obstacle.

[0051] The predicted driving trajectory may include multiple trajectory points, and the position information of each trajectory point may be represented by a coordinate point in an absolute coordinate system. The absolute coordinate system may be a coordinate system in which all coordinates are described based on the position of a fixed coordinate system origin, that is, each coordinate point in the absolute coordinate system does not vary depending on the reference object.

[0052] As a way, such as Figure 4 As shown, the method of obtaining the transformed path of the target obstacle based on the velocity potential field and the predicted driving trajectory of the target obstacle includes:

[0053] S131: Based on the position information of the multiple trajectory points and the velocity potential field, obtain the position information of the first path point corresponding to each of the multiple trajectory points.

[0054] As a method, if there is a trajectory point among multiple trajectory points that is located in the opposite direction of the driving direction corresponding to the velocity potential field, it is determined that the position information of the trajectory point is the same as the position information of the corresponding first path point; if there is a trajectory point among multiple trajectory points that is located in the same direction of the driving direction corresponding to the velocity potential field, based on the position information of the path point and the gradient of the velocity potential field, the position information of the first path point corresponding to the path point is obtained.

[0055] Among them, if the product of the target obstacle and the yaw angle corresponding to the velocity potential field is less than 0 ( The equation of a straight line is generated by this formula. The straight line corresponding to the equation can divide the velocity potential field into two parts. The opposite direction of the target obstacle's travel direction corresponding to the velocity potential field can be determined. If the product of the target obstacle and the travel direction corresponding to the velocity potential field is greater than or equal to 0, the positive direction of the target obstacle's travel direction corresponding to the velocity potential field can be determined.

[0056] The calculation formula for the position information of the first path point can be as follows:

[0057]

[0058] in, It can represent the position information of the trajectory point P0. It can express the gradient of the velocity potential field, (x i ,y i ) can represent the position information of the i-th remaining obstacle, ρ 0i It can represent the distance between the target obstacle and the i-th remaining obstacle, ρ i It can represent the repulsive force range of the i-th remaining obstacle, and η can represent the repulsive potential coefficient.

[0059] S132: Obtain the transformed path based on the position information of the first path point corresponding to each of the plurality of trajectory points.

[0060] As one approach, curve fitting may be performed on the position information of the plurality of first path points to obtain a transformed path.

[0061] For example, the transformation path can be as follows Figure 5 shown.

[0062] S140: Obtaining multiple differential entropies of the target obstacle based on the transformed path and multiple reference intended paths corresponding to the target obstacle, where the differential entropies represent similarities between the transformed path and the corresponding reference intended paths.

[0063] As a method, the path offset statistical results corresponding to the transformed path and multiple reference intention paths can be obtained to obtain multiple path offset statistical results corresponding to the target obstacle. The path offset statistical results can be used to characterize the difference between the transformed path and the corresponding intention reference path; based on multiple path offset statistical results, multiple differential entropies are obtained.

[0064] The reference intention path may refer to a reference path that does not violate traffic regulations under the lane where the target obstacle is located.

[0065] Optionally, a road connection relationship matrix can be obtained based on map information and navigation data. The road connection relationship matrix can represent the driving rules formed by every two lanes. The driving rules can be straight, left turn, right turn, or impassable. Based on the lane information of the target obstacle and the road connection relationship matrix, multiple reference intention paths are determined.

[0066] For example, Figure 6 As shown in the left figure, the current road can include 6 entrances (entering the intersection) and 6 exits (away from the intersection). The road connection relationship matrix can be as follows Figure 6 As shown in the right figure, 1 can represent a right turn, -1 can represent a left turn, 0 can represent a straight line, and ∞ can represent an impassable lane. When the target obstacle is located in the lane where entrance 1 is located, based on Figure 6 It can be seen that the reference intention path can be the path corresponding to exits 3, 4, and 5.

[0067] In an embodiment of the present application, the transformation path may include multiple first path points, and each reference intention path contains a second path point corresponding to each first path point.

[0068] Optional, such as Figure 7 As shown, the multiple path deviation statistics corresponding to the target obstacle are obtained based on the transformed path and the multiple reference intention paths, including:

[0069] S141: Based on the position information of the multiple first path points and the position information of the corresponding second path points in each of the reference intention paths, multiple reference distances of each of the reference intention paths are obtained, and the reference distance represents the distance between the corresponding first path point and the second path point.

[0070] As a method, for each first path point and a reference intention path, a reference distance representing the distance between the first path point and the second path point corresponding to the first path point in the reference intention path can be obtained based on the position information of the first path point and the position information of the second path point in the reference intention path, thereby obtaining multiple reference distances of the reference intention path.

[0071] Among them, the corresponding first path point and second path point can respectively refer to the intersection points of multiple horizontal lines with the predicted driving trajectory and the reference intention path.

[0072] Optionally, the distances between the multiple horizontal lines may be the same or different.

[0073] S142: Acquire a plurality of consecutive second path points having the same reference distance in each of the reference intention paths.

[0074] Exemplarily, the reference intention path may have consecutive second path point 1, second path point 2, second path point 3, second path point 4, and second path point 5, where the reference distance corresponding to second path point 1 may be 6 meters, the reference distance corresponding to second path point 2 may be 5 meters, the reference distances corresponding to second path point 3 and second path point 4 may both be 4 meters, and the reference distance corresponding to second path point 2 may be 3 meters. Then second path point 3 and second path point 4 are second path points with the same reference distance.

[0075] S143: Based on the multiple continuous second path points, obtain a sub-path in each reference intention path.

[0076] Exemplarily, the total length of the reference intention path can be 20 meters, and the distances between the second path point 1, the second path point 2, the second path point 3, the second path point 4, and the second path point 5 and the starting point of the reference intention path can be 1 meter, 5 meters, 10 meters, 15 meters, and 20 meters, respectively. The second path point 3 and the second path point 4 are second path points with the same reference distance. The sub-paths can include: a section of the reference intention path from the starting point of the reference intention path to the second path point 1, a section of the reference intention path from the second path point 1 to the second path point 2, a section of the reference intention path from the second path point 2 to the second path point 4, and a section of the reference intention path from the second path point 4 to the second path point 5.

[0077] S144: Obtain the plurality of path offset statistical results based on the length of the sub-path corresponding to each reference intended path and the reference distance corresponding to the length of the sub-path.

[0078] As a method, for each reference intention path, a segmented statistical function can be generated based on the length of the sub-path corresponding to the reference intention path and the reference distance corresponding to the length of the sub-path, and the segmented statistical function can be used as the preliminary path offset statistical result of the reference intention path. The preliminary path offset statistical result is then non-negativeized and regularized to obtain the path offset statistical result of the reference intention path.

[0079] Exemplarily, the subpaths may include: a reference intention path from the starting point of the reference intention path to the second path point 1, a reference intention path from the second path point 1 to the second path point 2, a reference intention path from the second path point 2 to the second path point 4, and a reference intention path from the second path point 4 to the second path point 5. The reference distance corresponding to the second path point 1 may be 6 meters, the reference distance corresponding to the second path point 2 may be 5 meters, the reference distances corresponding to the second path point 3 and the second path point 4 may both be 4 meters, and the reference distance corresponding to the second path point 2 may be 3 meters. The preliminary path deviation statistics can be expressed as:

[0080]

[0081] Optionally, the preliminary path deviation statistics are non-negativeized and regularized, and the formula for obtaining the path deviation statistics can be:

[0082]

[0083] Among them, e ε can represent the reference distance, σ can represent the value interval of the reference distance, γ(e ε ) can represent the preliminary path deviation statistics results.

[0084] Optionally, based on the statistical results of a single path deviation, the corresponding differential entropy calculation formula can be:

[0085]

[0086] Among them, e ε Can represent the reference distance, σ can represent the value interval of the reference distance, It can represent the path deviation statistics results.

[0087] S150: Based on a pre-trained hidden Markov model and the plurality of differential entropies, a driving intention of the target obstacle is obtained, where the driving intention is to go straight, turn left, or turn right.

[0088] The Hidden Markov Model (HMM) can be a generative model. The HMM model can be used to describe the dependency between two related sequences, which can be called a state sequence and an observation sequence. The state sequence can refer to an unobservable, hidden sequence, and the observation sequence can refer to an observable sequence. The value of the state sequence at time t is only related to the value of the state sequence at time t-1.

[0089] The HMM model can be determined by the initial probability vector (π), the state transition probability matrix (A) and the observation probability matrix (B). Among them, the state transition probability matrix can refer to the probability of transition between each state, such as: at time t in state q i Under the condition of j The probability of the observation probability matrix can refer to the probability of obtaining each observation value according to the current state, such as: at time t in state q i Generate observation v under the condition k The probability of the initial state is . The initial state probability vector can refer to the probability of each state appearing at the initial moment.

[0090] In an embodiment of the present application, multiple differential entropies may form a differential entropy sequence, the differential entropy sequence may be an observation sequence, and the driving intention sequence may be a state sequence.

[0091] As a method, the differential entropy sequence can be input into a pre-trained HMM model. Based on the Viterbi algorithm, a driving intention sequence corresponding to the differential entropy sequence can be obtained. Multiple driving intentions in the driving intention sequence can be converged based on their probabilities to obtain the driving intention of the target obstacle. Probability convergence can be understood as selecting the driving intention with the highest probability as the driving intention of the target obstacle.

[0092] In this embodiment of the present application, after obtaining multiple differential entropies representing the similarity between the transformed paths and the corresponding reference intention paths, these differential entropies are input into a pre-trained HMM model to obtain the target obstacle's driving intention. These differential entropies can then be further analyzed to improve the accuracy of the prediction results. Compared to methods that directly determine the target obstacle's driving intention based on multiple differential entropies, this method can improve the accuracy of the prediction results when the differences between the multiple differential entropies are not significant.

[0093] For example, Figure 8 As shown in the figure, when a target obstacle enters an intersection, its historical trajectory can be used to predict that its next trajectory may deviate to the left. At this point, the target obstacle's intention may be to turn left or to continue straight after avoiding the interactive obstacle. Therefore, after deriving multiple differential entropies, the target obstacle's final intention can be determined using a pre-trained hidden Markov model.

[0094] Optionally, the driving intention of the obstacles identified by the vehicle can be predicted one by one through steps S110 to S150, so as to change the speed, driving trajectory, etc. of the vehicle based on the prediction results, thereby improving the driving safety of the vehicle, especially the driving safety in scenarios where multiple vehicles are driving at unprotected intersections (intersections with no traffic controllers or traffic lights or malfunctioning traffic lights, etc.).

[0095] This embodiment provides a driving intention prediction method. After obtaining a target obstacle for a vehicle, the method constructs a velocity potential field corresponding to the target obstacle based on the distances between each of the remaining obstacles on the current road and the target obstacle, characterizing the impact of the remaining obstacle's speed on the target obstacle's route. A transformed path of the target obstacle is obtained based on the velocity potential field and the predicted driving trajectory of the target obstacle. Based on the transformed path and multiple reference intention paths corresponding to the target obstacle, multiple differential entropies of the target obstacle are obtained, characterizing the similarity between the transformed path and the corresponding reference intention paths. The driving intention of the target obstacle is determined based on a pre-trained hidden Markov model and the multiple differential entropies, which may indicate going straight, turning left, or turning right. Through the above method, a velocity potential field corresponding to the target obstacle can be constructed based on the distance between each of the remaining obstacles and the target obstacle, and the transformation path of the target obstacle can be obtained based on the velocity potential field and the predicted driving trajectory of the target obstacle. This can take into account the influence of the remaining obstacles on the driving route of the target obstacle, thereby improving the accuracy of the differential entropy obtained based on the transformation path and the reference intention path to characterize the similarity between the two. This makes it possible to obtain the driving intention of the target obstacle based on the accurate differential entropy and the pre-trained hidden Markov model, thereby improving the accuracy of the predicted driving intention of the target obstacle.

[0096] See also Figure 9 , an embodiment of the present application provides a driving intention prediction method, the method comprising:

[0097] S210: Acquire multiple pieces of training data, each piece of training data including position information of a target obstacle, a yaw angle, a true value of driving intention, obstacles that intersect with the predicted driving trajectory, and a passing order between obstacles that intersect with the predicted driving trajectory.

[0098] A true value of 1 indicates a right turn; a true value of -1 indicates a left turn; and a true value of 0 indicates going straight. Obstacles with intersecting predicted driving trajectories refer to obstacles whose predicted trajectory in the direction of travel overlaps with the vehicle's predicted trajectory, meaning the vehicle is traveling in the direction of the obstacle. The order of passage for obstacles with intersecting predicted driving trajectories can be yield or go first.

[0099] As a method, multiple training data as shown in Table 1 can be obtained through big data analysis and mining.

[0100] Table 1

[0101]

[0102] S220: Training the hidden Markov model to be trained based on the plurality of training data to obtain the hidden Markov model.

[0103] The parameters of the hidden Markov model to be trained may include an initial probability vector (π), a state transition probability matrix (A), and an observation probability matrix (B).

[0104] As one approach, maximum likelihood estimation may be performed based on multiple pieces of training data to obtain parameters of the hidden Markov model to be trained, thereby obtaining the hidden Markov model in step S150.

[0105] S230: Acquire a target obstacle of the vehicle.

[0106] S240: Constructing a velocity potential field corresponding to the target obstacle based on the distances between each of the remaining obstacles on the current road and the target obstacle, wherein the velocity potential field represents the influence of the travel speeds of the remaining obstacles on the travel route of the target obstacle.

[0107] S250: Obtaining a transformed path of the target obstacle based on the velocity resultant potential field and the predicted driving trajectory of the target obstacle.

[0108] S260: Obtaining multiple differential entropies of the target obstacle based on the transformed path and multiple reference intended paths corresponding to the target obstacle, where the differential entropies represent similarities between the transformed path and the corresponding reference intended paths.

[0109] S270: Based on a pre-trained hidden Markov model and the plurality of differential entropies, obtaining a driving intention of the target obstacle, where the driving intention is to go straight, turn left, or turn right.

[0110] This embodiment provides a method for predicting driving intention. Through the above-described method, a velocity potential field corresponding to a target obstacle is constructed based on the distances of each remaining obstacle from the target obstacle. Based on the velocity potential field and the target obstacle's predicted trajectory, a transformed path of the target obstacle is derived. This method takes into account the impact of the remaining obstacles on the target obstacle's path, thereby improving the accuracy of the differential entropy representing the similarity between the transformed path and the reference intention path. This allows the target obstacle's driving intention to be derived based on accurate differential entropy and a pre-trained hidden Markov model, thereby improving the accuracy of the predicted target obstacle's driving intention. Furthermore, in this embodiment, a hidden Markov model can be derived based on multiple training data consisting of a small number of data types, reducing the number of data types involved in training, thereby effectively reducing data dimensionality and model complexity, and improving the interpretability of the trained model. Compared to conventional methods that directly use target pose and historical trajectory data as the training set, the model training method in this application more rationally demarcates the boundary between artificial intelligence systems and rule-based systems. Moreover, by describing the similarity between driving intention and reference intention path through differential entropy, the rule system can be applied to analyze and process scene information to the maximum extent, avoiding the simple and crude output of initial data to the artificial intelligence system (hidden Markov model) for training, thereby reducing training costs and improving the accuracy and interpretability of prediction results.

[0111] See also Figure 10 The present application provides a driving intention prediction device 600, which includes:

[0112] The target obstacle acquisition unit 610 is used to acquire the target obstacle of the vehicle.

[0113] The transformed path acquisition unit 620 is configured to construct a velocity potential field corresponding to the target obstacle based on the distances between each of the remaining obstacles on the current road and the target obstacle. The velocity potential field represents the effect of the speeds of the remaining obstacles on the target obstacle's path. The transformed path acquisition unit 620 is configured to obtain the transformed path of the target obstacle based on the velocity potential field and the predicted trajectory of the target obstacle.

[0114] The differential entropy acquisition unit 630 is configured to obtain a plurality of differential entropies of the target obstacle based on the transformed path and a plurality of reference intention paths corresponding to the target obstacle, wherein the differential entropy represents a similarity between the transformed path and the corresponding reference intention path.

[0115] The driving intention obtaining unit 640 is configured to obtain the driving intention of the target obstacle based on a pre-trained hidden Markov model and the plurality of differential entropies, where the driving intention is to go straight, turn left, or turn right.

[0116] As one approach, the transformation path acquisition unit 620 is specifically configured to obtain the distance between each of the remaining obstacles and the target obstacle based on the yaw angle, size, speed of each of the remaining obstacles and the position information of the target obstacle; obtain the velocity potential field of each of the remaining obstacles based on the distance between each of the remaining obstacles and the target obstacle and the repulsive force range of each of the remaining obstacles; and obtain the combined velocity potential field based on the velocity potential field of each of the remaining obstacles.

[0117] As a method, the predicted driving trajectory includes multiple trajectory points, and the transformation path acquisition unit 620 is specifically used to obtain the position information of the first path point corresponding to each of the multiple trajectory points based on the position information of the multiple trajectory points and the velocity potential field; and obtain the transformation path based on the position information of the first path point corresponding to each of the multiple trajectory points.

[0118] Optionally, the transformation path acquisition unit 620 is specifically used to determine that if there is a trajectory point among the multiple trajectory points and the position information of the trajectory point is the same as the position information of the corresponding first path point, if there is a trajectory point among the multiple trajectory points and the position information of the trajectory point is located in the opposite direction of the driving direction corresponding to the velocity potential field; if there is a trajectory point among the multiple trajectory points and the position information of the first path point corresponding to the path point is obtained based on the position information of the path point and the gradient of the velocity potential field.

[0119] As a method, the differential entropy acquisition unit 630 is specifically used to obtain a road connection relationship matrix based on map information and navigation data, wherein the road connection relationship matrix represents the driving rules formed by every two lanes, and the driving rules are straight or left turn or right turn or impassable; based on the lane information of the target obstacle and the road connection relationship matrix, the multiple reference intention paths are determined.

[0120] As a method, the differential entropy acquisition unit 630 is specifically used to obtain the path deviation statistics corresponding to the transformed path and multiple reference intention paths, so as to obtain multiple path deviation statistics corresponding to the target obstacle, and the path deviation statistics are used to characterize the difference between the transformed path and the corresponding intended reference path; based on the multiple path deviation statistics, the multiple differential entropies are obtained.

[0121] Optionally, the transformed path includes multiple first path points, and each reference intention path contains a second path point corresponding to each first path point. The differential entropy acquisition unit 630 is specifically configured to obtain multiple reference distances for each reference intention path based on the position information of the multiple first path points and the position information of the corresponding second path point in each reference intention path, wherein the reference distance represents the spacing between the corresponding first path point and the second path point; obtain multiple consecutive second path points in each reference intention path that have the same reference distance; obtain a subpath in each reference intention path based on the multiple consecutive second path points; and obtain the multiple path offset statistical results based on the length of the subpath corresponding to each reference intention path and the reference distance corresponding to the subpath length.

[0122] The device 600 further includes:

[0123] The model training unit 640 is used to obtain multiple training data, each of which includes the position information, yaw angle, true value of driving intention, obstacles that intersect with the predicted driving trajectory, and the order of passage between obstacles that intersect with the predicted driving trajectory; and train the hidden Markov model to be trained based on the multiple training data to obtain the hidden Markov model.

[0124] The following will be combined Figure 11 A vehicle provided in this application is described.

[0125] See also Figure 11 Based on the aforementioned driving intention prediction method and apparatus, embodiments of the present application also provide another vehicle 100 capable of executing the aforementioned driving intention prediction method. Vehicle 100 includes one or more (only one shown in the figure) processors 102, a memory 104, and a data acquisition module 106, which are coupled to each other. The memory 104 stores a program capable of executing the contents of the aforementioned embodiments, and the processor 102 can execute the program stored in the memory 104.

[0126] The processor 102 may include one or more processing cores. The processor 102 utilizes various interfaces and circuits to connect various components within the vehicle 100. It executes instructions, programs, code sets, or instruction sets stored in the memory 104 and accesses data stored in the memory 104 to perform various functions and process data for the vehicle 100. Optionally, the processor 102 may be implemented in the form of at least one of a network processor (NPU), a digital signal processor (DSP), a field-programmable gate array (FPGA), or a programmable logic array (PLA). The processor 102 may integrate one or a combination of a central processing unit (CPU), a graphics processing unit (GPU), a network processor (NPU), and a modem. The CPU primarily processes the operating system, user interface, and application programs; the GPU is responsible for rendering and drawing display content; the NPU is responsible for processing multimedia data such as video and images; and the modem is responsible for wireless communication. It is understandable that the above-mentioned modem may not be integrated into the processor 102, but may be implemented separately through a communication chip.

[0127] The memory 104 may include random access memory (RAM) or read-only memory (ROM). The memory 104 may be used to store instructions, programs, codes, code sets, or instruction sets. The memory 104 may include a program storage area and a data storage area. The program storage area may store instructions for implementing an operating system, instructions for implementing at least one function (such as a touch function, a sound playback function, an image playback function, etc.), instructions for implementing the various method embodiments described below, and the like. The data storage area may also store data created by the vehicle 100 during use (such as a phone book, audio and video data, and chat history data).

[0128] The data acquisition module 106 can be used to obtain point cloud data, image data, etc. of the vehicle 100. The data acquisition module 106 can be a laser radar, a camera, a sensor, etc.

[0129] An embodiment of the present application provides a computer-readable storage medium 800. The computer-readable storage medium 800 stores program code, which can be called by a processor to execute the method described in the above method embodiment.

[0130] The computer-readable storage medium 800 can be an electronic memory such as a flash memory, an EEPROM (Electrically Erasable Programmable Read-Only Memory), an EPROM, a hard disk, or a ROM. Alternatively, the computer-readable storage medium 800 includes a non-transitory computer-readable storage medium. The computer-readable storage medium 800 has storage space for program code 810 for executing any of the method steps described above. These program codes can be read from or written to one or more computer program products. The program code 810 can be compressed, for example, in a suitable form.

[0131] In summary, the present application provides a driving intention prediction method, device, and vehicle. After obtaining the target obstacle of the vehicle, based on the distance between each of the remaining obstacles in the current road and the target obstacle, a speed potential field corresponding to the target obstacle is constructed, which characterizes the influence of the driving speed of the remaining obstacles on the driving route of the target obstacle. Based on the speed potential field and the predicted driving trajectory of the target obstacle, a transformation path of the target obstacle is obtained; based on the transformation path and multiple reference intention paths corresponding to the target obstacle, multiple differential entropies of the target obstacle are obtained, which characterize the similarity between the transformation path and the corresponding reference intention paths. Based on a pre-trained hidden Markov model and the multiple differential entropies, the driving intention of the target obstacle is obtained, and the driving intention is determined to be going straight, turning left, or turning right. Through the above method, a velocity potential field corresponding to the target obstacle can be constructed based on the distance between each of the remaining obstacles and the target obstacle, and the transformation path of the target obstacle can be obtained based on the velocity potential field and the predicted driving trajectory of the target obstacle. This can take into account the influence of the remaining obstacles on the driving route of the target obstacle, thereby improving the accuracy of the differential entropy obtained based on the transformation path and the reference intention path to characterize the similarity between the two. This makes it possible to obtain the driving intention of the target obstacle based on the accurate differential entropy and the pre-trained hidden Markov model, thereby improving the accuracy of the predicted driving intention of the target obstacle.

[0132] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A driving intention prediction method, characterized in that: The method comprises: Obtain the target obstacle of the vehicle; Based on the distances between each of the remaining obstacles on the current road and the target obstacle, constructing a velocity potential field corresponding to the target obstacle, wherein the velocity potential field represents the influence of the travel speeds of the remaining obstacles on the travel path of the target obstacle; Obtaining a transformed path of the target obstacle based on the velocity resultant potential field and the predicted driving trajectory of the target obstacle; obtaining a plurality of differential entropies of the target obstacle based on the transformed path and a plurality of reference intended paths corresponding to the target obstacle, wherein the differential entropies represent similarities between the transformed path and the corresponding reference intended paths; Based on a pre-trained hidden Markov model and the multiple differential entropies, a driving intention of the target obstacle is obtained, where the driving intention is to go straight, turn left, or turn right.

2. The method according to claim 1, characterized in that The constructing a velocity potential field corresponding to the target obstacle based on the distances between each of the remaining obstacles in the current road and the target obstacle includes: Obtaining a distance between each of the remaining obstacles and the target obstacle based on the yaw angle, size, and speed of each of the remaining obstacles and the position information of the target obstacle; Obtaining a velocity potential field of each of the remaining obstacles based on a distance between each of the remaining obstacles and the target obstacle and a repulsive force range of each of the remaining obstacles; The velocity resultant potential field is obtained based on the velocity potential field of each of the remaining obstacles.

3. The method according to claim 1, characterized in that The predicted driving trajectory includes a plurality of trajectory points. The predicted driving trajectory based on the velocity potential field and the target obstacle is used to obtain a transformed path of the target obstacle, including: Based on the position information of the plurality of trajectory points and the velocity resultant potential field, obtaining the position information of the first path point corresponding to each of the plurality of trajectory points; The transformed path is obtained based on the position information of the first path point corresponding to each of the multiple trajectory points.

4. The method according to claim 3, characterized in that The obtaining, based on the position information of the plurality of trajectory points and the velocity resultant potential field, position information of first path points corresponding to each of the plurality of trajectory points includes: If there is a trajectory point among the plurality of trajectory points that is located in a direction opposite to the traveling direction corresponding to the velocity resultant potential field, determining that the position information of the trajectory point is the same as the position information of the corresponding first path point; If there is a trajectory point among the multiple trajectory points located in the same direction as the driving direction corresponding to the velocity potential field, the position information of the first path point corresponding to the path point is obtained based on the position information of the path point and the gradient of the velocity potential field.

5. The method according to claim 1, wherein Before obtaining a plurality of differential entropies of the target obstacle based on the transformed path and a plurality of reference intended paths corresponding to the target obstacle, the method further includes: Obtaining a road connection relationship matrix based on map information and navigation data, wherein the road connection relationship matrix represents a driving rule formed by every two lanes, wherein the driving rule is straight ahead, left turn, right turn, or impassable. The multiple reference intention paths are determined based on the lane information of the target obstacle and the road connection relationship matrix.

6. The method according to claim 1, characterized in that The obtaining of a plurality of differential entropies of the target obstacle based on the transformed path and a plurality of reference intended paths corresponding to the target obstacle includes: Obtaining path deviation statistics corresponding to the transformed path and multiple reference intended paths, respectively, to obtain multiple path deviation statistics corresponding to the target obstacle, wherein the path deviation statistics are used to characterize the difference between the transformed path and the corresponding intended reference path; The multiple differential entropies are obtained based on the multiple path deviation statistical results.

7. The method according to claim 6, characterized in that The transformed path includes a plurality of first path points, each of the reference intention paths includes a second path point corresponding to each of the first path points, and obtaining a plurality of path deviation statistics corresponding to the target obstacle based on the transformed path and the plurality of reference intention paths includes: Based on the position information of the plurality of first path points and the position information of the corresponding second path point in each of the reference intended paths, obtaining a plurality of reference distances for each of the reference intended paths, the reference distances representing the spacing between the corresponding first path points and the second path points; Acquire a plurality of consecutive second path points having the same reference distance in each of the reference intended paths; Based on the multiple consecutive second path points, obtaining a subpath in each of the reference intention paths; The multiple path offset statistical results are obtained based on the length of the sub-path corresponding to each reference intention path and the reference distance corresponding to the length of the sub-path.

8. The method according to any one of claims 1 to 7, characterized in that: Before acquiring the target obstacle, the method further includes: Acquire a plurality of training data, each of the training data including position information of a target obstacle, a yaw angle, a true value of a driving intention, obstacles that intersect with the predicted driving trajectory, and a passing order between the obstacles that intersect with the predicted driving trajectory; The hidden Markov model to be trained is trained based on the multiple training data to obtain the hidden Markov model.

9. A driving intention prediction device, characterized in that: The device comprises: A target obstacle acquisition unit, used to acquire a target obstacle of the vehicle; a transformed path acquisition unit, configured to construct a velocity potential field corresponding to the target obstacle based on the distances between each of the remaining obstacles on the current road and the target obstacle, wherein the velocity potential field represents the effect of the speeds of the remaining obstacles on the path of the target obstacle; and to obtain a transformed path for the target obstacle based on the velocity potential field and the predicted trajectory of the target obstacle; a differential entropy acquisition unit, configured to obtain a plurality of differential entropies of the target obstacle based on the transformed path and a plurality of reference intended paths corresponding to the target obstacle, wherein the differential entropies represent a similarity between the transformed path and the corresponding reference intended paths; The driving intention obtaining unit is used to obtain the driving intention of the target obstacle based on a pre-trained hidden Markov model and the multiple differential entropies, where the driving intention is to go straight or turn left or turn right.

10. A vehicle, characterized in that: including one or more processors and memory; One or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs are configured to execute the method according to any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that The computer-readable storage medium stores program code, wherein when the program code is run, the method according to any one of claims 1 to 8 is executed.

Images