Methods and systems for creating reference image templates for identification of products on product storage structures of a product storage facility

Patent metadata

Field	Value
Publication number	US-2026044817-A1
Application number	US-202519361416-A
Country	US
Kind code	A1
Filing date	Oct 17, 2025
Priority date	Jan 24, 2023
Publication date	Feb 12, 2026
Grant date	—

Abstract

Systems and methods of creating reference template images for detecting and recognizing products at a product storage facility include an image capture device having a field of view that includes a product storage structure of the product storage facility, and a computing device including a control circuit communicatively coupled to the image capture device. The computing device obtains images of the product storage structure captured by the image capture device, analyzes the obtained images to detect individual products located on the product storage structure. Then, the computing device identifies the individual products detected in the images and crops each of the identified products from the images to generate cropped images. The computing device then creates a cluster of the cropped images and selects one of the cropped images as a reference template image of an identified individual product.

First claim

What is claimed is:

1 . A system of creating reference template images for detecting and recognizing products at a product storage structure of a product storage facility, the system comprising:
an image capture device configured to capture an image of the product storage structure; and a computing device comprising a control circuit, the computing device communicatively coupled to the image capture device, the control circuit configured to:
receive, from the computing device, directional movement instructions;
obtain, using a sensor communicatively coupled to the control circuit, based on the directional movement instructions received, a set of images of the product storage structure captured by the image capture device;
identify, from the set of images, a set of products;
associate each product of the set of products with a product identifier;
generate, from the set of images, a set of cropped images;
create a cluster from the set of cropped images, wherein each cropped image in the cluster depicts one of the products; and
select one image of the set of cropped images as a reference template image representing an associated product.

2 . The system of claim 1 , wherein the image capture device comprises a motorized robotic unit that includes wheels that permit the motorized robotic unit to move about the product storage facility, and a camera to permit the motorized robotic unit to capture the set of images of the product storage structure.

3 . The system of claim 1 , wherein the control circuit is programmed to generate virtual boundary lines for each of the obtained images, wherein each of the virtual boundary lines surrounds an individual one of the products captured in the obtained images.

4 . The system of claim 1 , wherein the control circuit is programmed to generate embeddings for each of the cropped images, wherein the embeddings represent dense vector representations of the cropped images.

5 . The system of claim 4 , wherein the control circuit is programmed to generate the embeddings for each of the cropped images using a convolutional neural network pretrained to extract predetermined features from the cropped images and to generate a lower dimensional representation of the cropped images.

6 . The system of claim 5 , wherein the control circuit is programmed to:
group the cropped images containing the embeddings into the cluster, wherein each of the cropped images in the cluster depicts one of the associated products; and select a centroid image of the cluster of the cropped images, wherein the centroid image is determined by the control circuit to represent a keyword template reference image of the one of the associated products.

7 . The system of claim 6 , wherein the control circuit is programmed to:
determine the embeddings of the cropped images in the cluster; and position the cropped images in the cluster based on the embeddings of the cropped images in the cluster.

8 . The system of claim 6 , wherein, the control circuit is programmed to:
resample a predetermined number of images of the cluster of the cropped images that are located closest to the centroid image; and mark the centroid image and the resampled images as feature vector template reference images of the one of the associated products.

9 . The system of claim 8 , the system further comprising an electronic database that stores the keyword template reference image and the feature vector template reference images associated with each one of the associated products to facilitate recognition of the products subsequently captured in at least one new image of the product storage structure by the image capture device.

10 . The system of claim 9 , wherein the control circuit is programmed to replace the one of the cropped images as the keyword template reference image or the feature vector template reference images of the one of the associated products in response to a determination by the control circuit that another cropped image obtained from the at least one new image represents the centroid image in an updated cluster of the cropped images of the one of the associated products.

11 . A method of creating reference template images for detecting and recognizing products at a product storage structure of a product storage facility, the method comprising:
by an image capture device, capturing an image of a product storage structure; and by a computing device including a control circuit, the control circuit communicatively coupled to the image capture device:
receiving, from the computing device, directional movement instructions;
obtaining, using a sensor communicatively coupled to the control circuit, based on the directional movement instructions received, a set of images of the product storage structure captured by the image capture device;
identifying, from the set of images, a set of products;
associating each product of the set of products with a product identifier;
generating, from the set of images, a set of cropped images;
creating a cluster from the set of cropped images, wherein each cropped image in the cluster depicts one of the products; and
selecting one image of the set of cropped images as a reference template image representing an associated product.

12 . The method of claim 11 , wherein the image capture device comprises a motorized robotic unit that includes wheels that permit the motorized robotic unit to move about the product storage facility, and a camera to permit the motorized robotic unit to capture the image of the product storage structure.

13 . The method of claim 11 , further comprising, by the control circuit, generating virtual boundary lines for each of the obtained images, wherein each of the virtual boundary lines surrounds an individual one of the products captured in the obtained images.

14 . The method of claim 11 , further comprising, by the control circuit, generating embeddings for each of the cropped images, wherein the embeddings represent dense vector representations of the cropped images.

15 . The method of claim 14 , further comprising, by the control circuit, generating the embeddings for each of the cropped images using a convolutional neural network pretrained to extract predetermined features from the cropped images and to generate a lower dimensional representation of the cropped images.

16 . The method of claim 15 , further comprising, by the control circuit:
grouping the cropped images containing the embeddings into the cluster, wherein each of the cropped images in the cluster depicts one of the associated products; and selecting a centroid image of the cluster of the cropped images, wherein the centroid image is determined by the control circuit to represent a keyword template reference image of the one of the associated products.

17 . The method of claim 16 , further comprising, by the control circuit:
determining the embeddings between the cropped images in the cluster; and positioning the cropped images in the cluster based on the embeddings of the cropped images in the cluster.

18 . The method of claim 16 , further comprising, by the control circuit:
resampling a predetermined number of images of the cluster of the cropped images that are located closest to the centroid image; and marking the centroid image and the resampled images as feature vector template reference images of the associated products.

19 . The method of claim 18 , further comprising storing the keyword template reference image and the feature vector template r

Assignees

• Walmart Apollo Llc

Inventors

• Ghosh Ashlin
• Balusu Raghava
• Pachauri Abhinav
• Jade Avinash M
• Zhang Lingfeng
• Jhunjhunwala Amit
• Robinson Jr William Craig
• Ellison Benjamin R
• Muktevi Srinivas
• Duan Zhaoliang

Classifications

• G06Q10/087 Primary

  Inventory or stock management, e.g. order filling, procurement or balancing against orders · CPC title