Global and local category-based item location filtering

1 . A system for category-based item location filtering to remove false positives, the system comprising:

an image capture device configured to generate image data within a retail facility;

a user interface device including a display;

a network enabling communication; and

a computer-readable medium storing instructions that are operative upon execution by a processor to:

receive, via the network, the image data generated by the image capture device;

determine a location in the retail facility from the image data using a location recognition model;

identify a plurality of candidate items associated with the determined location within a retail facility from the image data using an item recognition model;

remove false positive items associated with misplaced items within the retail facility from the plurality of candidate items by:

performing a global level filtering on the plurality of candidate items and removing any candidate items associated with an unconfirmed item category for the determined location using a location-to-category mapping table and an item-to-category mapping table; and

performing a local level filtering on the plurality of candidate items and removing low count items associated with the determined location and determined, from the image data, to have a count value below a majority threshold value;

generate an item-to-location mapping result comprising filtered items remaining in the plurality of candidate items following completion of the global level filtering and the local level filtering; and

present the item-to-location mapping result to a user via the display of the user interface device for accurate identification of item locations within the retail facility.

2 . The system of claim 1 , wherein the global level filtering further comprises:

identify a plurality of confirmed categories associated with the determined location using the location-to-category mapping table;

compare a category of a candidate item in the plurality of candidate items with the plurality of confirmed categories using the item-to-category mapping table;

generate a set of unconfirmed category items and a set of confirmed category items from the plurality of candidate items, wherein a confirmed category item in the set of confirmed category items comprises a candidate item associated with a confirmed category in the plurality of confirmed categories associated with the determined location; and

remove the set of unconfirmed category items from the plurality of candidate items, wherein an unconfirmed category item in the set of unconfirmed category items comprises a candidate item associated with a category absent from the plurality of confirmed categories.

3 . The system of claim 1 , wherein the local level filtering further comprises:

calculate an item count value for a selected item in the plurality of candidate items associated with the determined location;

compare the item count value for the selected item with the majority threshold value, wherein the selected item having an item count value which is less than the majority threshold is a low count item;

identify a set of item-level low count items and a set of item-level high count items, wherein a majority item in the set of item-level high count items is a candidate item having an item count at the determined location that is equal to or exceeds the majority threshold;

remove the set of item-level low count items from the plurality of candidate items; and

generate an item-to-location mapping for each unfiltered candidate item remaining in the plurality of candidate items for inclusion in a dynamic item-location map, wherein an unfiltered item is an item that is absent from a set of unconfirmed category items and the set of item-level low count items.

4 . The system of claim 3 , wherein the instructions are further operative to:

responsive to the set of item-level high count items being a null set, identify a set of category-level high count items and a set of category-level low count items based on item count values for all items in a same category at the determined location; and

remove candidate items in the set of category-level low count items from the plurality of candidate items.

5 . The system of claim 1 , wherein the instructions are further operative to:

validate location-to-category mapping for a selected location within the retail facility using historical data to confirm each category associated with the selected location.

6 . The system of claim 1 , wherein a single category is assigned to a selected item, wherein a single location is assigned a plurality of categories.

7 . The system of claim 1 , wherein the item-to-location mapping result comprises a plurality of entries, wherein each entry comprises a mapping of a recognized item to the determined location within the retail facility, and wherein the instructions are further operative to:

present the item-to-location mapping result to the user via the user interface device, wherein an entry associated with an item-out is moved to a first position on the user interface device for viewing by the user.

8 . A method for category-based item location filtering to remove false positives, the method comprising:

receiving, by a processor over a network, image data generated by an image capture device within a retail facility;

determining, by the processor, a location in the retail facility from the image data using a location recognition model;

identifying, by the processor, of a plurality of candidate items associated with the determined location from the image data using an item recognition model, wherein the item recognition model identifies the plurality of candidate items using an image generated by an image capture device;

removing, by the processor, false positive items associated with misplaced items within the retail facility from the plurality of candidate items by:

removing, by the processor using a global level filter, a first set of candidate items of the plurality of candidate items associated with an item category that is an unconfirmed category for the determined location, wherein the unconfirmed category is a category that is absent from a set of confirmed categories associated with the determined location; and

removing, by the processor, a second set of candidate items form the plurality of candidate items having an item-level low count within the determined location using a local level filter to reduce false positives in the plurality of candidate items, wherein items of the second set of candidate items are determined, from the image data, to have a count value that falls below a majority threshold value associated with the determined location;

generating, by the processor, an item-to-location mapping result comprising a mapping of filtered items remaining within the plurality of candidate items to the determined location; and

presenting, by the processor, the item-to-location mapping result to a user via a display of a user interface device for accurate identification of item locations within the retail facility.

9 . The method of claim 8 , further comprising:

identifying a set of category-level high count items and a set of category-level low count items based on item count values for all items in a same category at the determined location; and

removing candidate items in the set of category-level low count items from the plurality of candidate items.

10 . The method of claim 8 , wherein the false positives items are associated with misplaced items within the retail facility and items detected within shopping baskets.

11 . The method of claim 8 , further comprising:

generating image data associated with the determined location by an image capture device associated with a robotic device moving through the retail facility, wherein the item-to-location mapping result is updated dynamically in real-time as additional images of the determined location are received from the robotic device.

12 . The method of claim 8 , further comprising:

generating category-to-location mapping data for each category in a plurality of categories, wherein each category is assigned to at least one location within the retail facility.

13 . The method of claim 8 , further comprising:

generating a location-to-category mapping data for a selected location using planogram data, wherein the selected location is assigned a plurality of categories.

14 . The method of claim 8 , further comprising:

generating an item-to-category mapping for a selected item using item attribute data and per-category data, wherein a single category is assigned to the selected item.

15 . One or more computer storage devices having computer-executable instructions stored thereon, which, upon execution by a computer, are operative to:

receive, via a network, image data generated by an image capture device within a retail facility;

determine a location in the retail facility from the image data using a location recognition model;

identify a plurality of candidate items associated with determined location from the image data using an item recognition model, each candidate item in the plurality of candidate items comprising an item recognized by the item recognition model using an image of the candidate item and the determined location by the location recognition model;

remove false positive items associated with misplaced items within the retail facility from the plurality of candidate items by:

performing a global level filtering on the plurality of candidate items comprising identifying a set of unconfirmed category items and a set of confirmed category items in the plurality of candidate items using an item-to-category mapping table, an item in the set of confirmed category items comprising a candidate item associated with a category matching a confirmed category in a plurality of confirmed categories associated with the determined location;

removing the set of unconfirmed category items from the plurality of candidate items, wherein each item in the set of unconfirmed category items comprises a candidate item associated with a category absent from the plurality of confirmed categories;

performing a local level filtering on the set of confirmed category items, the local level filtering comprising identifying a set of item-level low count items determined, from the image data, to have an item count value within the determined location less than a majority threshold value;

removing the set of item-level low count items from the plurality of candidate items;

map, by an item-to-location mapping, a set of candidate items remaining in the plurality of candidate items following global level filtering and local level filtering to remove false positives associated with the item recognition models and the location recognition model; and

present the item-to-location mapping to a user via a display of a user interface device for accurate identification of item locations within the retail facility.

16 . The one or more computer storage devices of claim 15 , wherein the operations further comprise:

identify a set of item-level high count items and a set of item-level low count items based on item count values for all recognized items at the determined location; and

remove the candidate items in the set of item-level low count items from the plurality of candidate items.

17 . The one or more computer storage devices of claim 15 , wherein the operations further comprise:

apply zero bin tolerance metrics and one bin tolerance metrics to item-to-location mapping results.

18 . The one or more computer storage devices of claim 15 , wherein the operations further comprise:

generate a location-to-category mapping for a selected location within the retail facility using historical data and planogram data associated with a plurality of sections within the retail facility.

19 . The one or more computer storage devices of claim 15 , wherein the operations further comprise:

responsive to a set of item-level high count items being a null set, identify a set of category-level high count items and a set of category-level low count items based on item count values for all items in a same category at the determined location; and

remove candidate items in the set of category-level low count items from the plurality of candidate items.

20 . The one or more computer storage devices of claim 15 , wherein the operations further comprise:

generate an item-to-location mapping result having false positives filtered out for improved accuracy, a location-to-item mapping result comprising a plurality of entries, wherein each entry comprises a mapping of a recognized item to the determined location within the retail facility; and

present the item-to-location mapping result to a user via a user interface device, wherein an entry associated with an item-out is moved to a first position on the user interface device for viewing by the user.