System and method for detecting and classifying recurrent stops of a vehicle fleet

1. A method for identifying recurrent stops of a vehicle fleet having a plurality of vehicles, comprising:

retrieving, by a device, historical GPS tracks of the vehicle fleet over a period of time,

wherein the device comprises one or more processors;

detecting, by the device, stops made by the vehicle fleet along travelled routes that are associated with the historical GPS tracks;

constructing, by the device, a coverage area that covers the travelled routes;

discretizing, by the device, the coverage area into a plurality of cells;

determining, by the device, whether a cell is a first recurrent stop based on a fleet stay period associated with the cell,

wherein the fleet stay period is determined by:

tallying a quantity of total stops by the plurality of vehicles in the cell, and

calculating a fleet stay period from a total length of time of the plurality of vehicles stopped in the cell;

determining, by the device, whether an adjacent cell, of the plurality of cells, is a second recurrent stop based on a fleet stay period associated with the adjacent cell;

merging, by the device, the cell with the adjacent cell to form a merged single recurrent stop from the first recurrent stop and the second recurrent stop; and

classifying, by the device, the merged single recurrent stop into a plurality of categories.

2. The method of claim 1 , wherein the period of time includes a plurality of work shifts.

3. The method of claim 1 , wherein the cell has a size of about 100 meters.

4. The method of claim 1 , further comprising:

determining whether a vehicle associated with a retrieved GPS track is in a STOP state, an IDLING state, or a JOURNEY state.

5. The method of claim 4 , wherein the fleet stay period of a cell is a total length of time when any vehicle of the vehicle fleet shows up in the cell in a STOP state or an IDLING state.

6. The method of claim 1 , wherein the classifying implements a random forest algorithm to classify a recurrent stop into a depot, an employee home, or other locations.

7. The method of claim 6 , wherein features used by the random forest algorithm include:

an average number of stops per day;

mean and standard deviations of stop durations;

an average cumulative stop duration per day; and

a percentage of overnight stops.

8. The method of claim 7 , wherein the features further include:

an aspect ratio of a bounding box of a recurrent stop;

a percentage of fleet vehicles that stop at least once in the recurrent stop;

maximum and minimum stop durations;

a mean density of stops per day; and

a percentage of stops staring or ending in a specific time of the day.

9. The method of claim 1 , further comprising obtaining a satellite image corresponding to a recurrent stop,

wherein the classifying implements a convolutional neural network that classifies the recurrent stop into the plurality of categories based on the satellite image.

10. The method of claim 1 ,

wherein the classifying combines a random forest algorithm and a convolutional neural network to classify a recurrent stop,

the random forest algorithm using a plurality of computed features to classify the recurrent stop, and

the convolutional neural network using a satellite image to classify the recurrent stop, and

wherein a classification result of the convolutional neural network is included in the plurality of computed features used by the random forest algorithm.

11. A non-transitory storage medium storing an executable program that, when executed by a processor, causes the processor to:

retrieve historical GPS tracks of a vehicle fleet, comprising a plurality of vehicles, over a period of time;

detect stops made by the vehicle fleet along travelled routes that are associated with the historical GPS tracks;

construct a coverage area that covers the travelled routes;

discretize the coverage area into a plurality of cells;

determine whether a cell is a first recurrent stop based on a fleet stay period associated with the cell,

wherein the fleet stay period is determined by:

tallying a quantity of total stops by the plurality of vehicles in the cell, and

calculating the fleet stay period from a total length of time of the plurality of vehicles stopped in the cell;

determine whether an adjacent cell, of the plurality of cells, is a second recurrent stop based on a fleet stay period associated with the adjacent cell;

merge the cell with the adjacent cell to form a merged single recurrent stop from the first recurrent stop and the second recurrent stop; and

classify the merged single recurrent stop into a plurality of categories.

12. The non-transitory storage medium of claim 11 , wherein the period of time includes a plurality of work shifts.

13. The non-transitory storage medium of claim 11 , wherein the cell has a size of about 100 meters.

14. The non-transitory storage medium of claim 11 , wherein the executable program further causes the processor to:

determine whether a vehicle associated with a retrieved GPS track is in a STOP state, an IDLING state, or a JOURNEY state.

15. The non-transitory storage medium of claim 14 , wherein the fleet stay period of a cell is a total length of time when any vehicle of the vehicle fleet shows up in the cell in a STOP state or an IDLING state.

16. The non-transitory storage medium of claim 11 , wherein the executable program, when causing the processor to classify, causes the processor to:

implement a random forest algorithm to classify a recurrent stop into a depot, an employee home, or other locations.

17. The non-transitory storage medium of claim 16 , wherein features used by the random forest algorithm include:

an average number of stops per day;

mean and standard deviations of stop durations;

an average cumulative stop duration per day; and

a percentage of overnight stops.

18. The non-transitory storage medium of claim 17 , wherein the features further include:

an aspect ratio of a bounding box of a recurrent stop;

a percentage of fleet vehicles that stop at least once in the recurrent stop;

maximum and minimum stop durations;

a mean density of stops per day; and

a percentage of stops staring or ending in a specific time of the day.

19. The non-transitory storage medium of claim 11 , wherein the executable program further causes the processor to:

obtain a satellite image corresponding to a recurrent stop,

wherein, the executable program, when causing the processor to classify, causes the processor to:

implement a convolutional neural network that classifies the recurrent stop into the plurality of categories based on the satellite image.

20. A device comprising:

a memory; and

a processor, communicatively coupled to the memory, to:

retrieve historical GPS tracks of a vehicle fleet, comprising a plurality of vehicles, over a period of time,

detect stops made by the vehicle fleet along travelled routes that are associated with the historical GPS tracks,

construct a coverage area that covers the travelled routes,

discretize the coverage area into a plurality of cells,

determine whether a cell is a first recurrent stop based on a fleet stay period associated with the cell,

wherein the fleet stay period is determined by:

tallying a quantity of total stops by the plurality of vehicles in the cell, and

calculating the fleet stay period from a total length of time of the plurality of vehicles stopped in the cell;

determine whether an adjacent cell, of the plurality of cells, is a second recurrent stop based on a fleet stay period associated with the adjacent cell;

merge the cell with the adjacent cell to form a merged single recurrent stop from the first recurrent stop and the second recurrent stop, and

classify a the merged single recurrent stop into a plurality of categories.