Method and apparatus for driver intent prediction

1 . A method for predicting a driver intent of a driver, the method comprising:

receiving driving environment information comprising lane information and trajectory information of a vehicle;

receiving driving information of the vehicle;

generating a lane feature vector by inputting the lane information into a first encoder implemented as a first graph convolutional network;

generating a trajectory feature vector by inputting the trajectory information into a second encoder implemented as a second graph convolutional network;

generating a driving feature vector by inputting the driving information into a third encoder implemented as an artificial neural network; and

predicting the driver intent by inputting at least one of the lane feature vector, the trajectory feature vector or the driving feature vector into a driver intent prediction model, wherein the driver intent is represented as a destination grid cell on a grid map corresponding to a drivable region of the vehicle.

2 . The method of claim 1 , further comprising training the driver intent prediction model by using the driver intent and the trajectory information.

3 . The method of claim 2 , wherein training the driver intent prediction model comprises:

labeling the trajectory information;

generating a predicted driving route by using the driver intent;

comparing the labeled trajectory information with the predicted driving route; and

changing a weight of the driver intent prediction model based on a result of the comparing.

4 . The method of claim 1 , wherein generating outputting the driving feature vector comprises:

pre-processing the driving information by removing noise from the driving information; and

inputting the pre-processed driving information into the third encoder.

5 . The method of claim 1 , further comprising generating the grid map by using the lane information and the driving information.

6 . The method of claim 5 , wherein generating the grid map comprises:

determining a size of a region for which the grid map is to be generated by using the driving information and the lane information; and

generating the grid map by dividing the region into a plurality of grids.

7 . The method of claim 6 , wherein generating the grid map by dividing the region into the plurality of grids comprises dividing the region such that each grid of the plurality of grids in a driving lane of the vehicle has a different size from other grids of the plurality of grids on other lanes.

8 . The method of claim 1 , wherein predicting the driver intent comprises concatenating and inputting the lane feature vector, the trajectory feature vector and the driving feature vector into the driver intent prediction model.

9 . The method of claim 1 , wherein inputting at least one of the lane feature vector, the trajectory feature vector or the driving feature vector into the driver intent prediction model further comprises inputting traffic signal information included in the driving environment information into the driver intent prediction model.

10 . The method of claim 1 , wherein:

generating the lane feature vector comprises:

normalizing the lane information by removing noise from the lane information and representing the normalized lane information as a first graph; and

inputting the normalized lane information into the first encoder; and

generating the trajectory feature vector comprises:

normalizing the trajectory information by removing noise from the trajectory information and representing the normalized trajectory information as a second graph; and

inputting the normalized trajectory information into the second encoder.

11 . The method of claim 1 , further comprising implementing driving safety control based upon the predicted driver intent.

12 . An apparatus comprising:

one or more processors; and

a non-transitory storage device storing instructions that, when executed by the one or more processors, cause the one or more processors to:

receive driving environment information comprising lane information and trajectory information of a vehicle;

receive driving information of the vehicle;

generate a lane feature vector by inputting the lane information into a first encoder implemented as a graph convolutional network;

generate a trajectory feature vector by inputting the trajectory information into a second encoder implemented as a graph convolutional network;

generate a driving feature vector by inputting the driving information into a third encoder implemented as an artificial neural network; and

predict a driver intent by inputting at least one of the lane feature vector, the trajectory feature vector or the driving feature vector into a driver intent prediction model, wherein the driver intent is represented as a destination grid cell on a grid map corresponding to a drivable region of the vehicle and wherein driving safety control can be implemented based on the predicted driver intent.

13 . The apparatus of claim 12 , wherein the instructions further cause the one or more processors to train the driver intent prediction model by using the driver intent and the trajectory information.

14 . The apparatus of claim 13 , wherein the instructions further cause the one or more processors to:

generate a predicted driving route by using the driver intent;

label the trajectory information;

compare the labeled trajectory information with the predicted driving route; and

change a weight of the driver intent prediction model based on a result of the comparison.

15 . The apparatus of claim 12 , wherein the instructions further cause the one or more processors to:

determine a size of a region for which the grid map is to be generated by using the driving information and the lane information; and

generate the grid map by dividing the region into a plurality of grids.

16 . The apparatus of claim 12 , wherein the instructions further cause the one or more processors to:

generate normalized lane information and normalized trajectory information by representing the lane information and the trajectory information as a graph, respectively;

generate the lane feature vector by inputting the normalized lane information into the first encoder; and

generate the trajectory feature vector by inputting the normalized trajectory information into the second encoder.

17 . The apparatus of claim 12 , wherein the instructions further cause the one or more processors to predict the driver intent by inputting the at least one of the lane feature vector, the trajectory feature vector or the driving feature vector and traffic signal information included in the driving environment information into the driver intent prediction model.

18 . A method for predicting a driver intent of a driver in a vehicle, the method comprising:

receiving driving environment information comprising lane information, trajectory information, and traffic signal information;

receiving driving information of the vehicle comprising at least one of an acceleration, a yaw rate, a steering angle, or a speed;

pre-processing the lane information by removing noise and representing the lane information as a first graph comprising nodes connected by edges;

pre-processing the trajectory information by removing noise and representing the trajectory information as a second graph comprising nodes connected by edges;

pre-processing the driving information by removing noise using a low-pass filter;

generating a grid map by determining a size of a drivable region using the speed and acceleration from the driving information and a width of a driving lane from the lane information and dividing the drivable region into a plurality of grids wherein each grid in a current driving lane of the vehicle has a different size than grids in adjacent lanes;

generating a lane feature vector by inputting the pre-processed lane information into a first encoder implemented as a first graph convolutional network;

generating a trajectory feature vector by inputting the pre-processed trajectory information into a second encoder implemented as a second graph convolutional network;

generating a driving feature vector by inputting the pre-processed driving information into a third encoder implemented as an artificial neural network;

concatenating the lane feature vector, the trajectory feature vector, and the driving feature vector to form a combined feature vector;

predicting the driver intent by inputting the combined feature vector and the traffic signal information into a driver intent prediction model implemented as a deep neural network, wherein the predicted driver intent is represented as a destination grid cell on the grid map; and

implementing driving safety control based upon the predicted driver intent.

19 . The method of claim 18 , wherein the trajectory information comprises only trajectory data within a predetermined distance from a current location of the vehicle.

20 . The method of claim 18 , wherein dividing the drivable region into the plurality of grids comprises making a size of each grid in the current driving lane 0.5 times a size of each grid in the adjacent lanes.