Vehicle driving intention prediction method, apparatus, and terminal, and storage medium

1. A method comprising:

obtaining map information of an intersection when a target vehicle drives to the intersection, wherein the map information of the intersection comprises road layer information and lane layer information, wherein the road layer information is of roads connected to the intersection, and wherein the lane layer information is of lanes of the roads;

obtaining driving information of the target vehicle;

determining a first driving feature of the target vehicle relative to each of the lanes based on the driving information and the lane layer information;

determining a second driving feature of the target vehicle relative to each of the roads based on the driving information and the road layer information;

determining a lane intention of the target vehicle based on a third driving feature of a surrounding vehicle relative to the target vehicle, the first driving feature, and the second driving feature, wherein the surrounding vehicle is in a preset range of the target vehicle, and wherein the lane intention represents a first probability distribution that the target vehicle drives away from the intersection through each of the lanes;

determining a driving intention of the target vehicle based on the lane intention; and

controlling, by a control system of the target vehicle, the target vehicle to drive on a planned driving route determined based on the driving intention.

2. The method of claim 1 , wherein determining the lane intention comprises determining a road intention of the target vehicle and the lane intention based on the first driving feature, the second driving feature, and the third driving feature, wherein the road intention represents a second probability distribution that the target vehicle drives away from the intersection through each of the roads, and wherein determining the driving intention comprises determining the driving intention based on the road intention and the lane intention.

3. The method of claim 2 , wherein determining the road intention comprises:

inputting a fourth driving feature of the surrounding vehicle into a first interaction feature extraction network to determine an interaction feature vector between the surrounding vehicle and the target vehicle, wherein the interaction feature vector represents an impact of the surrounding vehicle on the target vehicle;

inputting the interaction feature vector, the first driving feature, and a fifth driving feature of the target vehicle relative to a road associated with each of the plurality of lanes into a lane intention prediction network to determine the lane intention and a first driving feature implicit vector of the target vehicle relative to each of the lanes; and

inputting the second driving feature, the first driving feature implicit vector that is of the target vehicle relative to each of the lanes and that corresponds to the lanes associated with each of the roads, and the lane intention corresponding to a lane associated with each of the roads into a road intention prediction network to determine the road intention.

4. The method of claim 3 , wherein the first interaction feature extraction network comprises a plurality of first feature extraction subnetworks and an interaction feature vector prediction network, and wherein inputting the fourth driving feature comprises:

inputting the fourth driving feature into the first feature extraction subnetworks to determine a first driving feature vector of the surrounding vehicle; and

inputting the first driving feature vector into the interaction feature vector prediction network to determine the interaction feature vector.

5. The method of claim 3 , wherein the lane intention prediction network comprises a first lane feature extraction subnetwork, a second lane feature extraction subnetwork, a first road feature extraction subnetwork, a second road feature extraction subnetwork, and a lane intention prediction subnetwork, and wherein inputting the interaction feature vector, the first driving feature, and the fifth driving feature into the lane intention prediction network comprises:

inputting the first driving feature into the first lane feature extraction subnetwork to determine a second driving feature vector of the target vehicle relative to each of the lanes;

inputting the second driving feature vector into the second lane feature extraction subnetwork to determine the first driving feature implicit vector;

inputting the fifth driving feature into the first road feature extraction subnetwork to determine a third driving feature vector of the target vehicle relative to the road associated with each of the lanes;

inputting the third driving feature vector into the second road feature extraction subnetwork to determine a second driving feature implicit vector of the target vehicle relative to the road associated with each of the lanes; and

inputting the interaction feature vector, the second driving feature vector, the first driving feature implicit vector, and the second driving feature implicit vector into the lane intention prediction subnetwork to determine the lane intention.

6. The method of claim 5 , wherein inputting the second driving feature vector comprises:

extracting, by a plurality of feature extraction windows in the second lane feature extraction subnetwork, the second driving feature vector in an order of driving moments; and

determining the first driving feature implicit vector based on an implicit vector output by a previous feature extraction window and a fourth driving feature vector that is of the target vehicle relative to each of the lanes and that corresponds to a current feature extraction window.

7. The method of claim 5 , wherein inputting the third driving feature vector comprises:

extracting, by a plurality of feature extraction windows in the second road feature extraction subnetwork, the third driving feature vector in an order of driving moments; and

determining the second driving feature implicit vector based on an implicit vector output by a previous feature extraction window and a fifth driving feature vector that is of the target vehicle relative to the road associated to each of the lanes and that corresponds to a current feature extraction window.

8. The method of claim 5 , wherein the road intention prediction network comprises the first road feature extraction subnetwork, the second road feature extraction subnetwork, and a road intention prediction subnetwork, and wherein inputting the second driving feature, the first driving feature implicit vector, and the lane intention into the road intention prediction network comprises:

inputting the second driving feature into the first road feature extraction subnetwork to determine a sixth driving feature vector of the target vehicle relative to each of the roads;

inputting the sixth driving feature vector into the second road feature extraction subnetwork to determine a third driving feature implicit vector of the target vehicle relative to each of the roads;

performing, based on the lane intention corresponding to the lane, weighted processing on the first driving feature implicit vector;

splicing weighted processing results to obtain an implicit fusion vector of the lane; and

inputting the second driving feature vector and the implicit fusion vector of the lane into the road intention prediction subnetwork to determine the road intention.

9. The method of claim 8 , wherein the first road feature extraction subnetwork is based on at least a multi-layer perceptron network and a recurrent neural network, wherein the first lane feature extraction subnetwork and the first road feature extraction subnetwork are based on at least the multi-layer perceptron network, wherein the second lane feature extraction subnetwork and the second road feature extraction subnetwork are based on at least the recurrent neural network, and wherein an interaction feature vector prediction network, the lane intention prediction subnetwork, and the road intention prediction subnetwork are all based on at least attention in neural networks.

10. The method of claim 2 , wherein determining the driving intention comprises:

determining a road corresponding to a maximum probability in the road intention as a target road;

determining a lane corresponding to a maximum probability in the lane intention corresponding to a lane associated with the target road as a target lane; and

determining the driving intention based on the target road and the target lane.

11. The method of claim 1 , wherein the third driving feature comprises one or more of a location feature of the surrounding vehicle in a first coordinate system, a speed feature, or a head orientation feature, wherein an origin of the first coordinate system is a current location of the target vehicle, wherein the first coordinate system is a rectangular coordinate system, wherein a y-axis of the first coordinate system is parallel to a length direction of a vehicle body of the target vehicle, and wherein a forward direction of the y-axis is consistent with a head orientation of the target vehicle.

12. The method of claim 1 , wherein the second driving feature comprises one or more of a location feature of the target vehicle in a second coordinate system, a distance feature between the target vehicle and an origin, a head orientation feature of the target vehicle, or a feature including that a location of the target vehicle in the second coordinate system, a distance between the target vehicle and the origin, and a head orientation of the target vehicle change with a driving moment, wherein the second coordinate system is a rectangular coordinate system, wherein an origin of the second coordinate system is based on an exit location of each of the roads, and wherein an x-axis direction is based on a driving direction of each of the roads.

13. The method of claim 1 , wherein the first driving feature comprises one or more of a location feature of the target vehicle in a third coordinate system, a first feature of an angle formed by a head orientation of the target vehicle and a driving direction of a lane, or a second feature including that a location of the target vehicle in the third coordinate system and the angle change with a driving moment, wherein the third coordinate system is a frenet coordinate system, wherein a reference line of the third coordinate system is based on a center line of each of the lanes, and wherein an origin of the third coordinate system is based on an end point of the center line.

14. The method of claim 1 , further comprising determining the lanes based on a topology analysis on the roads.

15. An apparatus comprising:

a memory configured to store instructions; and

at least one processor coupled to the memory and configured to:

obtain map information of an intersection when a target vehicle drives to the intersection, wherein the map information of the intersection comprises road layer information and lane layer information, wherein the road layer information is of roads connected to the intersection, and wherein the lane layer information is of lanes of the roads;

obtain driving information of the target vehicle;

determine a first driving feature of the target vehicle relative to each of the lanes based on the driving information and the lane layer information;

determine a second driving feature of the target vehicle relative to each of the roads based on the driving information and the road layer information;

determine a lane intention of the target vehicle based on a third driving feature of a surrounding vehicle relative to the target vehicle, the first driving feature, and the second driving feature, wherein the surrounding vehicle is in a preset range of the target vehicle, and wherein the lane intention represents a first probability distribution that the target vehicle drives away from the intersection through each of the plurality of lanes;

determine a driving intention of the target vehicle based on the lane intention; and

control, by a control system of the target vehicle, the target vehicle to drive on a planned driving route determined based on the driving intention.

16. The apparatus of claim 15 , wherein the at least one processor is further configured to:

determine a road intention of the target vehicle and the lane intention based on the first driving feature, the second driving feature, and the third driving feature, wherein the road intention of the target vehicle represents a second probability distribution that the target vehicle drives away from the intersection through each of the roads; and

determine the driving intention based on the road intention and the lane intention.

17. The apparatus of claim 16 , wherein the at least one processor is further configured to:

input a fourth driving feature of the surrounding vehicle into a first interaction feature extraction network to determine an interaction feature vector between the surrounding vehicle and the target vehicle, wherein the interaction feature vector represents an impact of the surrounding vehicle on the target vehicle;

input the interaction feature vector, the first driving feature, and a fifth driving feature of the target vehicle relative to a road associated with each of the lanes into a lane intention prediction network to determine the lane intention and a first driving feature implicit vector of the target vehicle relative to each of the lanes; and

input the second driving feature, the first driving feature implicit vector that is of the target vehicle relative to each of the plurality of lanes and that corresponds to the lanes associated with each of the roads, and the lane intention corresponding to a lane associated with each of the roads into a road intention prediction network to determine the road intention.

18. The apparatus of claim 17 , wherein the first interaction feature extraction network comprises a plurality of first feature extraction subnetworks and an interaction feature vector prediction network, and wherein the at least one processor is further configured to:

input the fourth driving feature into the first feature extraction subnetworks to determine a first driving feature vector of the surrounding vehicle; and

input the first driving feature vector into the interaction feature vector prediction network to determine the interaction feature vector.

19. The apparatus of claim 17 , wherein the lane intention prediction network comprises a first lane feature extraction subnetwork, a second lane feature extraction subnetwork, a first road feature extraction subnetwork, a second road feature extraction subnetwork, and a lane intention prediction subnetwork, and wherein the at least one processor is further configured to:

input the first driving feature into the first lane feature extraction subnetwork to determine a second driving feature vector of the target vehicle relative to each of the lanes;

input the second driving feature vector into the second lane feature extraction subnetwork to determine the first driving feature implicit vector;

input the fifth driving feature into the first road feature extraction subnetwork to determine a third driving feature vector of the target vehicle relative to the road;

input the third driving feature vector into the second road feature extraction subnetwork to determine a second driving feature implicit vector of the target vehicle relative to the road associated with each of the lanes; and

input the interaction feature vector, the second driving feature vector, the first driving feature implicit vector, and the second driving feature implicit vector into the lane intention prediction subnetwork to determine the lane intention.

20. A computer program product comprising computer-executable instructions stored on a non-transitory computer-readable storage medium that, when executed by a processor, cause an apparatus to:

obtain map information of an intersection when a target vehicle drives to the intersection, wherein the map information of the intersection comprises road layer information and lane layer information, wherein the road layer information is of roads connected to the intersection, and wherein the lane layer information is of lanes of the roads;

obtain driving information of the target vehicle;

determine a first driving feature of the target vehicle relative to each of the lanes based on the driving information and the lane layer information;

determine a second driving feature of the target vehicle relative to each of the roads based on the driving information and the road layer information;

determine a lane intention of the target vehicle based on a third driving feature of a surrounding vehicle relative to the target vehicle, the first driving feature, and the second driving feature, wherein the surrounding vehicle is in a preset range of the target vehicle, and wherein the lane intention represents a probability distribution that the target vehicle drives away from the intersection through each of the plurality of lanes;

determine a driving intention of the target vehicle based on the lane intention; and

control, by a control system of the target vehicle, the target vehicle to drive on a planned driving route determined based on the driving intention.