Vehicle classification using a recurrent neural network (RNN)

• US 10,345,449 B2
• Filed: 05/22/2017
• Issued: 07/09/2019
• Est. Priority Date: 12/02/2016
• Status: Active Grant

• Current Assignee: Verizon Connect Development Limited
• Original Assignee: Verizon Connect Ireland Limited (Verizon Communications Inc.)
• Inventors: Salti, Samuele, Sambo, Francesco, Taccari, Leonardo, Bravi, Luca, Simoncini, Matteo, Lori, Alessandro
• Primary Examiner(s): Black, Thomas G
• Assistant Examiner(s): Lewandroski, Sara J
• Application Number:
  US15/601,814
• Publication Number:
  US 20180157963A1
• Time in Patent Office: 778 Days
• CPC Class Codes: G01S 19/14 specially adapted for specific applications
• CPC Class Codes: G05B 13/027 using neural networks only
• CPC Class Codes: G05B 2219/21002 Neural classifier for inputs, groups inputs into classes
• CPC Class Codes: G06N 3/02 Neural networks
• CPC Class Codes: G06N 3/044 Recurrent networks, e.g. Hopfield networks
• CPC Class Codes: G06N 3/045 Combinations of networks
• CPC Class Codes: G08G 1/0129 for creating historical data or processing based on historical data

Abstract

A device can receive GPS data or values for a set of metrics at a set of GPS points that form a GPS track of a vehicle. The device can determine additional values for additional metrics using the GPS data or the values for the set of metrics. The device can determine a set of vectors for the set of GPS points using the GPS data, the values, or the additional values. The set of vectors can be used in a recurrent neural network (RNN) to classify the vehicle. The device can process the set of vectors via one or more sets of RNN layers of the RNN. The device can determine a classification of the vehicle using a result of processing the set of vectors. The result can be output by the output layer. The device can perform an action based on the classification of the vehicle.

Claims

1. A device, comprising:

one or more processors to;

receive global positioning system (GPS) data and values for a set of metrics at a set of GPS points that form a GPS track of a vehicle;

determine additional values for additional metrics using the GPS data or the values for the set of metrics;

determine a set of vectors for the set of GPS points using the GPS data, the values, or the additional values,

the set of vectors to be used in a recurrent neural network (RNN) to classify the vehicle;

process the set of vectors via one or more sets of RNN layers of the RNN,

the one or more sets of RNN layers including;

one or more input layers,

multiple sets of feed forward layers,

a set of recurrent layers, and

an output layer;

determine a classification of the vehicle using a result of processing the set of vectors,

the result being output by the output layer; and

perform an action based on the classification of the vehicle.

2. The device of claim 1, where the one or more processors are further to:

train the RNN using a training set of vectors prior to receiving the GPS data or the values for the set of metrics.

3. The device of claim 1, where the one or more processors are further to:

combine the GPS data, the values, or the additional values to generate a numerical representation of the GPS data, the values, or the additional values; and

where the one or more processors, when determining the set of vectors, are to;

determine the set of vectors based on combining the GPS data, the values, or the additional values.

4. The device of claim 1, where the one or more processors, when receiving the GPS data or the values, are to:

receive the GPS data or the values at a low frequency interval.

5. The device of claim 1, where the one or more processors, when determining the additional values, are to:

determine a road type on which the vehicle is travelling using the GPS data or the values.

6. The device of claim 1, where the one or more processors, when determining the classification, are to:

determine the classification using a fixed-length vector output by the output layer.

7. The device of claim 1, where the one or more processors, when performing the action, are to:

generate a report or an alert based on the classification of the vehicle,

the report or the alert being related to fuel consumption or driving style of the vehicle.

8. A non-transitory computer-readable medium storing instructions, the instructions comprising:

one or more instructions that, when executed by one or more processors, cause the one or more processors to;

receive global positioning system (GPS) data or values for a set of metrics at a set of GPS points that form a GPS track of a vehicle;

determine additional values for additional metrics using the GPS data or the values for the set of metrics;

determine a set of vectors for the set of GPS points using the GPS data, the values, or the additional values,

the set of vectors to be used in a recurrent neural network (RNN) to classify the vehicle;

process the set of vectors via one or more sets of RNN layers of the RNN,

the one or more sets of RNN layers including;

one or more input layers,

multiple sets of feed forward layers,

a set of recurrent layers, and

an output layer;

determine a classification of the vehicle using a result of processing the set of vectors,

the result being output by the output layer; and

perform an action based on the classification of the vehicle.

9. The non-transitory computer-readable medium of claim 8, where the one or more instructions, that cause the one or more processors to process the set of vectors, cause the one or more processors to:

process a first vector, of the set of vectors, via a first set of RNN layers, of the one or more sets of RNN layers, that includes;

an input layer, of the one or more input layers,

a first set of feed forward layers, of the multiple sets of feed forward layers, and

a first set of recurrent layers, of the set of recurrent layers; and

provide the first vector to a second set of recurrent layers, of the set of recurrent layers, included in a second set of RNN layers.

10. The non-transitory computer-readable medium of claim 9, where the one or more instructions, when executed by the one or more processors, further cause the one or more processors to:

process the first vector and a second vector, of the set of vectors, via the second set of recurrent layers;

provide the result of processing the first vector and the second vector to a second set of feed forward layers, of the multiple sets of feed forward layers, included in the second set of RNN layers to further process the result; and

output the result via the output layer after processing the first vector and the second vector using the second set of feed forward layers.

11. The non-transitory computer-readable medium of claim 8, where the set of recurrent layers is a set of long short-term memory (LSTM) layers.

12. The non-transitory computer-readable medium of claim 8, where the one or more sets of RNN layers include a pooling layer.

13. The non-transitory computer-readable medium of claim 8, where the result is a fixed-length vector.

14. The non-transitory computer-readable medium of claim 8, where the one or more instructions, that cause the one or more processors to perform the action, cause the one or more processors to:

generate a report that includes information identifying;

the classification of the vehicle, or

a quantity of vehicles associated with the classification.

15. A method, comprising:

receiving, by a device, global positioning system (GPS) data or values for a set of metrics at a set of GPS points that form a GPS track of a vehicle;

determining, by the device, additional values for additional metrics using the GPS data or the values for the set of metrics;

determining, by the device, a set of vectors for the set of GPS points using the GPS data, the values, or the additional values,

the set of vectors to be used in a recurrent neural network (RNN) to classify the vehicle;

processing, by the device, the set of vectors via one or more sets of RNN layers of the RNN,

the one or more sets of RNN layers including;

one or more input layers,

multiple sets of feed forward layers,

a set of recurrent layers, and

an output layer;

determining, by the device, a classification of the vehicle using a result of processing the set of vectors,

the result being output by the output layer; and

performing, by the device, an action based on the classification of the vehicle.

16. The method of claim 15, where the multiple sets of feed forward layers include:

a set of fully-connected pre-processing layers, and

a set of fully-connected post-processing layers.

17. The method of claim 15, further comprising:

modifying a weight that the one or more sets of RNN layers use to process the set of vectors during training of the RNN.

18. The method of claim 15, where the one or more sets of RNN layers include a pooling layer that implements a pooling function,

the pooling function including:

max pooling, or

average pooling.

19. The method of claim 15, where determining the classification comprises:

determining the classification using information identifying a probability that the vehicle is one or more classifications,

the one or more classifications including the classification.

20. The method of claim 15, where determining the classification comprises:

determining the classification based on an average result of processing the set of vectors multiple times using the RNN.