Vehicle occupancy detection using passenger to driver feature distance

1. A method for vehicle occupancy detection comprising:

collecting image data of a vehicle;

creating a first value according to a plurality of characteristics of a driver position of said vehicle;

creating at least one other value according to a plurality of characteristics of at least one other candidate occupant position in said vehicle;

calculating a distance between said first value and said at least one other value in order to determine a vehicle occupancy;

determining a threshold distance based on a training set of images wherein said threshold distance is indicative of a passenger in a vehicle; and

applying a classifier to said distance between said value and said at least one other value in order to determine if said distance between said first value and said at least one other value is indicative of a passenger in said vehicle.

2. The method of claim 1 wherein collecting image data of a vehicle further comprises:

localizing said image data of said vehicle in order to define a localized windshield region; and

splitting said localized windshield region into a driver part and a passenger part.

3. The method of claim 2 wherein said image data of a vehicle further comprises near infrared image data of said vehicle.

4. The method of claim 2 wherein creating said first value according to a plurality of characteristics of a driver position of said vehicle further comprises calculating a first image feature representation vector; and

wherein creating at least one other value according to a plurality of characteristics of at least one other candidate occupant position in said vehicle further comprises calculating at least one other image feature representation vector.

5. The method of claim 4 wherein calculating a first image feature representation vector and calculating at least one other image feature representation vector further comprises:

extracting a set of patches;

computing a set of statistics from each of said patches; and

aggregating said set of statistics from each of said patches into at least one of an image-level representation and a region level representation.

6. The method of claim 4 wherein comparing said first image feature representation vector to said at least one other image feature representation vector further comprises calculating a distance between said first image feature representation vector and said at least one other image feature representation vector.

7. The method of claim 6 wherein said distance is calculated using at least one of:

a Euclidean distance; and

a Mahalanobis distance.

8. The method of claim 1 wherein said classifier comprises one of:

a kernel machine; and

a probabilistic clustering algorithm.

9. A method for vehicle occupancy detection comprising:

collecting image data of a vehicle;

localizing said image data of said vehicle in order to define a localized windshield region;

splitting said localized windshield region into a driver part and a passenger part;

creating a first value according to a plurality of characteristics of said driver part of said vehicle by calculating a first image feature representation vector;

creating at least one other value according to a plurality of characteristics of said passenger part of said vehicle by calculating at least one other image feature representation vector;

calculating a distance between said first image feature representation vector and said at least one other image feature representation vector in order to determine a vehicle occupancy;

determining a threshold distance based on a training set of images wherein said threshold distance is indicative of a passenger in a vehicle; and

applying a classifier to said distance between said first image feature representation vector and said at least one other image feature representation vector in order to determine if said distance between said first image feature representation vector and said at least one other image feature representation vector is indicative of a passenger in said vehicle.

10. A system for vehicle occupancy detection comprising:

a camera module configured to collect image data of a vehicle;

a processor; and

a non-transitory computer-usable medium embodying computer code, said computer-usable medium being coupled to said processor, said computer code comprising instructions executable by said processor configured for:

collecting image data of a vehicle;

creating a first value according to a plurality of characteristics of a driver position of said vehicle;

creating at least one other value according to a plurality of characteristics of at least one other candidate occupant position in said vehicle;

calculating a distance between said first value and said at least one other value in order to determine a vehicle occupancy;

determining a threshold distance based on a training set of images wherein said threshold distance is indicative of a passenger in a vehicle; and

applying a classifier to said distance between said first value and said at least one other value in order to determine if said distance between said first value and said at least one other value is indicative of a passenger in said vehicle.

11. The system of claim 10 wherein collecting image data of a vehicle further comprises:

localizing said image data of said vehicle in order to define a localized windshield region; and

splitting said localized windshield region into a driver part and a passenger part.

12. The system of claim 10 wherein said image data of a vehicle further comprises near infrared image data of said vehicle.

13. The system of claim 11 wherein creating said first value according to a plurality of characteristics of a driver position of said vehicle further comprises calculating a first image feature representation vector; and

wherein creating at least one other value according to a plurality of characteristics of at least one other candidate occupant position in said vehicle further comprises calculating at least one other image feature representation vector.

14. The system of claim 13 wherein calculating a first image feature representation vector and calculating at least one other image feature representation vector further comprises:

extracting a set of patches;

computing a set of statistics from each of, said patches; and

aggregating said set of statistics from each of said patches into at least one of an image-level representation and a region level representation.

15. The system of claim 14 wherein comparing said first image feature representation vector to said at least one other image feature representation vector further comprises calculating a distance between said first image feature representation vector and said at least one other image feature representation vector.

16. The system of claim 15 wherein said distance is calculated using at least one of:

a Euclidean distance; and

a Mahalanobis distance.

17. The system of claim 10 wherein said classifier comprises one of:

a kernel machine; and

a probabilistic clustering algorithm.