System and method for point-to-point traffic prediction

1. A method for point-to-point traffic prediction, comprising:

obtaining, by a processor and from a plurality of computing devices, time-series locations of a plurality of vehicles respectively associated with the computing devices, wherein: the time-series locations form first trajectory data comprising corresponding trajectories at least passing from a first point O to a second point D within a first time interval;

obtaining, by a detector, a traffic volume between O and D for a second time interval that is temporally after the first time interval;

training, by the processor, one or more weights of a neural network model by inputting the first trajectory data and the traffic volume to the neural network model and using the obtained traffic volume as ground truth to obtain a trained neural network model, wherein the training one or more weights of the neural network model further comprises minimizing a total loss, and the total loss comprises a loss parameter associated with a difference between the obtained traffic volume and a predicted traffic volume predicted based on the first trajectory data; and

inputting, by the processor, second trajectory data between O and D to the trained neural network model to predict a future traffic volume between O and D for the a future time interval.

2. The method of claim 1 , wherein:

the traffic volume comprises a number of vehicles that travel from O to D for the second time interval; and

obtaining the traffic volume comprises:

capturing first images of vehicle plates of a first number of all vehicles passing through O within the second time interval, and capturing second images of vehicle plates of a second number of all vehicles passing through D within the second time interval t; and

determining a number of vehicles with vehicle plates detected from the first images and subsequently from the second images as the number of vehicles.

3. The method of claim 1 , further comprising:

in response to determining the predicted future traffic volume to exceed a threshold, sending a signal, by the processor, to a traffic signal controller to allow more traffic to pass through during the future time interval.

4. The method of claim 1 , wherein:

the neural network model comprises: an input layer, a first BiLSTM (bidirectional long short-term memory) layer, a first BN (batch normalization) layer, a second BiLSTM layer, a second BN layer, a first dense layer, a third BN layer, a second dense layer, and an output layer connected in series.

5. A method for point-to-point traffic prediction, comprising:

obtaining, by a processor and from a plurality of computing devices, time-series locations of a plurality of vehicles respectively associated with the computing devices;

extracting, from the time-series locations, at least (1) trajectory data i comprising corresponding trajectories at least passing through a first point pair within a first time interval and (2) trajectory data j comprising corresponding trajectories at least passing through a second point pair within the first time interval, wherein:

the first point pair and the second point pair each comprise a point O and a point D, traffic flowing from the point O to the point D; and

the first point pair and the second point pair have at least one of: different O points or different D points;

obtaining, by a detector, a traffic volume between the first point pair for a second time interval that is temporally after the first time interval;

training, by the processor, one or more weights of a neural network model by inputting the trajectory data i, the trajectory data j, and the traffic volume to the neural network model to obtain a trained neural network model, wherein the neural network model comprises a correlation between the trajectory data i and the trajectory data j; and

inputting, by the processor, trajectory data k between the second point pair to the trained neural network model to predict a future traffic volume through the second point pair for the a future time interval.

6. The method of claim 5 , wherein:

the traffic volume comprises a number of vehicles that travel through the first point pair for the second time interval; and

obtaining the traffic volume comprises:

capturing first images of vehicle plates of a first number of all vehicles passing through the O point of the first point pair within the second time interval, and capturing second images of vehicle plates of a second number of all vehicles passing through the D point of the first point pair within the second time interval; and

determining a number of vehicles with vehicle plates detected from the first images and subsequently from the second images as the number of vehicles.

7. The method of claim 5 , further comprising:

in response to determining the predicted future traffic volume to exceed a threshold, sending a signal, by the processor, to a traffic signal controller to allow more traffic to pass through during the future time interval.

8. The method of claim 5 , wherein:

the neural network model comprises: an input layer, a first LSTM (long short-term memory) layer, a first BN (batch normalization) layer, a second LSTM layer, a second BN layer, a dense layer, and an output layer connected in series.

9. The method of claim 8 , wherein:

the first BN layer is further outputted to a third LSTM layer;

the third LSTM layer is outputted to a third BN layer; and

the third BN layer is outputted to another output layer for obtaining projected trajectory data.

10. The method of claim 9 , wherein:

training the one or more weights of the neural network model comprises minimizing a total loss; and

the total loss comprises a first loss parameter associated with a difference between the trajectory data i and the projected trajectory data.

11. The method of claim 5 , wherein:

training the one or more weights of the neural network model comprises minimizing a total loss; and

the total loss comprises a second loss parameter associated with the correlation.

12. The method of claim 11 , wherein:

the second loss parameter is further associated with a difference between a predicted traffic volume for the first point pair and a predicted traffic volume for the second point pair; and

the predicted traffic volume for the second point pair is adjusted by a ratio trained by a random forest model.

13. The method of claim 5 , wherein:

training the one or more weights of the neural network model comprises minimizing a total loss;

the total loss comprises a third loss parameter associated with a difference between a predicted traffic volume and the obtained traffic volume; and

the predicted traffic volume is predicted based on the trajectory data i.

14. A system for point-to-point traffic prediction, comprising:

a detector configured to obtain a traffic volume between a first point pair for a second time interval that is temporally after a first time interval; and

a processor configured to:

obtain the traffic volume from the detector, and obtain, from a plurality of computing devices, time-series locations of a plurality of vehicles respectively associated with the computing devices;

extract, from the time-series locations, at least (1) trajectory data i comprising corresponding trajectories at least passing through the first point pair within the first time interval and (2) trajectory data j comprising corresponding trajectories at least passing through a second point pair within the first time interval, wherein:

the first point pair and the second point pair each comprise a point O and a point D, traffic flowing from the point O to the point D; and

the first point pair and the second point pair have at least one of: different O points or different D points;

train one or more weights of a neural network model by inputting the trajectory data i, the trajectory data j, and the traffic volume to the neural network model to obtain a trained neural network model, wherein the neural network model comprises a correlation between the trajectory data i and the trajectory data j; and

input trajectory data k between the second point pair to the trained neural network model to predict a future traffic volume through the second point pair for the a future time interval.

15. The system of claim 14 , wherein:

the traffic volume comprises a number of vehicles that travel through the first point pair for the second time interval; and

to obtain the traffic volume, the processor is configured to:

capture first images of vehicle plates of a first number of all vehicles passing through the O point of the first point pair within the second time interval, and capture second images of vehicle plates of a second number of all vehicles passing through the D point of the first point pair within the second time interval; and

determine a number of vehicles with vehicle plates detected from the first images and subsequently from the second images as the number of vehicles.

16. The system of claim 14 , wherein the processor is further configured to:

in response to determining the predicted future traffic volume to exceed a threshold, send a signal to a traffic signal controller to allow more traffic to pass through during the future time interval.

17. The system of claim 14 , wherein:

the neural network model comprises: an input layer, a first LSTM (long short-term memory) layer, a first BN (batch normalization) layer, a second LSTM layer, a second BN layer, a dense layer, and an output layer connected in series;

the first BN layer is further outputted to a third LSTM layer;

the third LSTM layer is outputted to a third BN layer; and

the third BN layer is outputted to another output layer for obtaining projected trajectory data.

18. The system of claim 17 , wherein:

to train the one or more weights of the neural network model, the processor is configured to minimize a total loss; and

the total loss comprises a first loss parameter associated with a difference between the trajectory data i and the projected trajectory data.

19. The system of claim 14 , wherein:

to train the one or more weights of the neural network model, the processor is configured to minimize a total loss;

the total loss comprises a second loss parameter associated with the correlation and with a difference between a predicted traffic volume for the first point pair and a predicted traffic volume for the second point pair; and

the predicted traffic volume for the second point pair is adjusted by a ratio trained by a random forest model.

20. The system of claim 14 , wherein:

to train the one or more weights of the neural network model, the processor is configured to minimize a total loss;

the total loss comprises a third loss parameter associated with a difference between a predicted traffic volume and the obtained traffic volume; and

the predicted traffic volume is predicted based on the trajectory data i.