Barrier detection for support structures

1. A method of barrier detection in an imaging controller, the method comprising:

obtaining, by the imaging controller from an image sensor, an image of a support structure configured to support a plurality of items on a support surface extending between a shelf edge and a shelf back;

extracting, by the imaging controller, frequency components representing pixels of the image;

based on the extracted frequency components, identifying, by the imaging controller, a barrier region of the image, the barrier region containing a barrier adjacent to the shelf edge; and

detecting, by the imaging controller, at least one empty sub-region within the barrier region, wherein the empty sub-region is free of items between the barrier and the shelf back.

2. The method of claim 1 , further comprising:

obtaining an item region defining a location of an item within the image; and

generating a corrected item region based on the item region and the empty sub-region.

3. The method of claim 2 , wherein generating the corrected item region includes:

detecting a portion of the empty sub-region that overlaps with the item region; and

subtracting the overlapping portion from the item region.

4. The method of claim 3 , further comprising:

prior to the subtracting, comparing a first area of the overlapping portion to a second area of the item region;

when the first area represents a smaller fraction of the second area than a lower threshold, ignoring the overlapping portion; and

when the first area represents a greater fraction of the second area than an upper threshold, discarding the item region.

5. The method of claim 1 , wherein extracting the set of frequency components includes:

for each row of pixels in the image, generating a frequency-domain representation of the row; and

selecting a set of frequency components from the frequency-domain representation.

6. The method of claim 5 , wherein identifying the barrier region includes:

assigning each row of pixels in the image one of (i) a barrier class, and (ii) a non-barrier class, based on classification of the corresponding set of frequency components;

positioning a selection window at an upper edge of the image;

determining whether a portion of the rows of pixels within the window having the barrier class exceeds a threshold; and

when the determination is negative, shifting the window towards a lower edge of the image, and repeating the determination.

7. The method of claim 6 , wherein the classification of the corresponding set of frequency components includes providing the set of frequency components to a support vector machine (SVM).

8. The method of claim 6 , further comprising:

when the determination is affirmative, setting (i) an upper edge of the barrier region based on the position of the window.

9. The method of claim 1 , wherein detecting the empty sub-region includes:

for each pixel within the barrier region:

generating a set of features, and

based on the set of features, generating a probability that the pixel represents an empty region;

dividing the image into a series of bins; and

setting each bin as empty when the mean probability of the pixels within the bin exceeds a threshold.

10. The method of claim 9 , wherein generating the probability includes providing the set of features to a classifier.

11. A computing device, comprising:

a memory;

an image sensor; and

an imaging controller configured to:

obtain, from the image sensor, an image of a support structure configured to support a plurality of items on a support surface extending between a shelf edge and a shelf back;

extract frequency components representing pixels of the image;

based on the extracted frequency components, identify a barrier region of the image, the barrier region containing a barrier adjacent to the shelf edge; and

detect at least one empty sub-region within the barrier region, wherein the empty sub-region is free of items between the barrier and the shelf back.

12. The computing device of claim 11 , wherein the imaging controller is further configured to:

obtain an item region defining a location of an item within the image; and

generate a corrected item region based on the item region and the empty sub-region.

13. The computing device of claim 12 , wherein the imaging controller is configured, to generate the corrected item region, to:

detect a portion of the empty sub-region that overlaps with the item region; and

subtract the overlapping portion from the item region.

14. The computing device of claim 13 , wherein the imaging controller is further configured to:

prior to the subtraction, compare a first area of the overlapping portion to a second area of the item region;

when the first area represents a smaller fraction of the second area than a lower threshold, ignore the overlapping portion; and

when the first area represents a greater fraction of the second area than an upper threshold, discard the item region.

15. The computing device of claim 11 , wherein the imaging controller is configured, to extract the set of frequency components, to:

for each row of pixels in the image, generate a frequency-domain representation of the row; and

select a set of frequency components from the frequency-domain representation.

16. The computing device of claim 15 , wherein the imaging controller is configured, to identify the barrier region, to:

assign each row of pixels in the image one of (i) a barrier class, and (ii) a non-barrier class, based on classification of the corresponding set of frequency components;

position a selection window at an upper edge of the image;

determine whether a portion of the rows of pixels within the window having the barrier class exceeds a threshold; and

when the determination is negative, shift the window towards a lower edge of the image, and repeat the determination.

17. The computing device of claim 16 , wherein the imaging controller is configured, to generate the classification of the corresponding set of frequency components, to provide the set of frequency components to a support vector machine (SVM).

18. The computing device of claim 16 , wherein the imaging controller is further configured to:

when the determination is affirmative, set (i) an upper edge of the barrier region based on the position of the window.

19. The computing device of claim 11 , wherein the imaging controller is configured, to detect the empty sub-region, to:

for each pixel within the barrier region:

generate a set of features, and

based on the set of features, generate a probability that the pixel represents an empty region;

divide the image into a series of bins; and

set each bin as empty when the mean probability of the pixels within the bin exceeds a threshold.

20. The computing device of claim 19 , wherein the imaging controller is configured, to generate the probability, to provide the set of features to a random forest classifier.