Vehicle recognition apparatus and vehicle recognition method

1. A vehicle recognition apparatus comprising:

one or more hardware processors to

acquire a photographed image obtained by photographing, using a stereo camera, a plurality of vehicles forming a column of vehicles on a road from an obliquely rearward direction with respect to the column of vehicles,

generate a parallax image based on the photographed image,

estimate a vanishing point in the parallax image,

generate a plurality of strip regions by dividing a region below the vanishing point in the parallax image in a horizontal direction,

calculate a representative parallax for each of the strip regions,

extract strip regions from which the plurality of vehicles is recognizable, from among the strip regions based on the representative parallax,

couple strip regions adjacent to each other among the extract strip regions to generate a coupled strip region based on a distance distribution model for a vehicular side and back surfaces, and

recognize the plurality of vehicles from the coupled strip region.

2. The vehicle recognition apparatus according to claim 1 , wherein the one or more hardware processors

determine, based on a number of the strip regions in the coupled strip region, whether the coupled strip region is a division target,

calculate top two parallaxes having higher frequency of occurrence in each of the strip regions constituting the coupled strip region that is determined to be the division target,

calculate a difference in distances based on the top two parallaxes,

determine, when the calculated difference in distances is included in a predetermined range, a horizontal position in the strip region to be a horizontal position for dividing the coupled strip region into two regions, and

divide the coupled strip region into two divided regions based on the determined horizontal position,

wherein the recognition of one of the plurality of vehicles is performed based on each of the two divided regions.

3. The vehicle recognition apparatus according to claim 1 , wherein the one or more hardware processors determine a number of strip regions based on resolution of the parallax image, and divide, based on the determined number, the region below the vanishing point in the parallax image in the horizontal direction to generate the strip regions.

4. The vehicle recognition apparatus according to claim 1 , wherein the one or more hardware processors calculate a frequency distribution of parallax having a parallax value equal to or larger than a first predetermined value among parallaxes of respective pixels for each of the strip regions, and set a most frequent value of the frequency distribution as the representative parallax.

5. The vehicle recognition apparatus according to claim 4 , wherein the one or more hardware processors extract, from the strip regions in which the representative parallax exists, a cluster of pixels having the representative parallax as the strip region from which one or more of the plurality of vehicles is recognizable.

6. The vehicle recognition apparatus according to claim 4 , wherein, when there are two clusters of pixels having the representative parallax and a difference in distances between lower end positions of the respective two clusters is equal to or larger than a second predetermined value, the one or more hardware processors extract a larger cluster of the two clusters as the strip region from which one or more of the plurality of vehicles is recognizable.

7. The vehicle recognition apparatus according to claim 1 , wherein the distance distribution model represents that a distance to each point on the vehicular back surface is substantially constant, and that a distance to each point on the vehicular side surface along a traveling direction of the road varies within a predetermined variation rate.

8. The vehicle recognition apparatus according to claim 7 , wherein the one or more hardware processors convert the representative parallax of each of the strip regions into a distance based on the distance distribution model, and couple two adjacent strip regions to generate the coupled strip region when a distance of the two adjacent strip regions is equal to or smaller than a third predetermined value.

9. The vehicle recognition apparatus according to claim 2 , wherein the one or more hardware processors determine the coupled strip region as the division target when the number of the strip regions in the coupled strip region is equal to or larger than a fourth predetermined value.

10. The vehicle recognition apparatus according to claim 9 , wherein the one or more hardware processors calculate a frequency distribution of parallax for each of the strip regions constituting a region on a side on which a side surface of one of the plurality of vehicles is reflected with respect to center in the horizontal direction of the coupled strip region determined as the division target, and calculate the top two parallaxes having higher frequency of occurrence.

11. The vehicle recognition apparatus according to claim 10 , wherein, when each frequency of occurrence of each of the top two parallaxes is equal to or larger than a fifth predetermined value, the one or more hardware processors convert the top two parallax into distances, and calculate the difference in the distances.

12. The vehicle recognition apparatus according to claim 11 , wherein

the one or more hardware processors determine whether the calculated difference in distances is included in the predetermined range for each of the strip regions in a direction from the center in the horizontal direction of the coupled strip region determined as the division target toward a side on which the side surface of the vehicle is reflected, and,

when the difference in distances is included in the predetermined range, the one or more hardware processors determine the horizontal position in the strip region as the horizontal position for dividing the coupled strip region into two regions.

13. The vehicle recognition apparatus according to claim 1 , wherein the one or more hardware processors

create, based on the parallax image, bird's-eye view data representing a view of the road from directly above, and

detect a rear end part of one of the plurality of vehicles based on the bird's-eye view data.

14. The vehicle recognition apparatus according to claim 13 , wherein the one or more hardware processors create the bird's-eye view data based on pieces of parallax data at points same in the horizontal direction of the road and different in height, using the parallax image.

15. The vehicle recognition apparatus according to claim 13 , wherein the one or more hardware processors extract a boundary between a fence and the one of the plurality of vehicles based on the bird's-eye view data, and detect the rear end part of the vehicle one of the plurality of vehicles.

16. A vehicle recognition method comprising:

acquiring a photographed image obtained by photographing, using a stereo camera, a plurality of vehicles forming a column of vehicles on a road from an obliquely rearward direction with respect to the column of vehicles;

generating a parallax image based on the photographed image;

estimating a vanishing point in the parallax image;

generating a plurality of strip regions by dividing a region below the vanishing point in the parallax image in a horizontal direction;

calculating a representative parallax for each of the strip regions;

extracting strip regions from which the plurality of vehicles is recognizable, from among the strip regions based on the representative parallax,

coupling strip regions adjacent to each other among the extract strip regions to generate a coupled strip region based on a distance distribution model for a vehicular side and back surfaces, and

recognizing the plurality of vehicles from the coupled strip region.

17. The vehicle recognition method according to claim 16 , further comprising:

determining, based on a number of the strip regions in the coupled strip region, whether the coupled strip region is a division target;

calculating top two parallaxes having higher frequency of occurrence in each of the strip regions constituting the coupled strip region that is determined to be the division target;

calculating a difference in distances based on the top two parallaxes;

determining, when the calculated difference in distances is included in a predetermined range, a horizontal position in the strip region to be a horizontal position for dividing the coupled strip region into two regions; and

dividing the coupled strip region into two divided regions based on the determined horizontal position,

wherein the recognizing the plurality of vehicles comprises recognizing the plurality of vehicles from each of the two divided regions.

18. The vehicle recognition method according to claim 16 , wherein the generating a plurality of strip regions comprises determining a number of strip regions based on resolution of the parallax image, and dividing, based on the determined number, the region below the vanishing point in the parallax image in the horizontal direction to generate the strip regions.

19. The vehicle recognition method according to claim 16 , wherein the calculating a representative parallax comprises calculating a frequency distribution of parallax having a parallax value equal to or larger than a first predetermined value among parallaxes of respective pixels for each of the strip regions, and setting a most frequent value of the frequency distribution as the representative parallax.

20. The vehicle recognition method according to claim 19 , wherein the extracting strip regions comprises extracting, from the strip regions in which the representative parallax exists, a cluster of pixels having the representative parallax as the strip region from which one or more of the plurality of vehicles is recognizable.