System and method for identifying products in a shelf management system
Disclosed herein is a system and method of identifying products on a retail shelf using a feature extractor trained to extract features from images of products on the shelf and output identifying information regarding the product in the product image. The extracted features are compared to extracted features in a feature gallery library and a best fit match is obtained. A product ID is then assigned to the image of the product and the assigned product ID is validated by matching the product ID with product identifying information extracted from a shelf label associated with the image of the product.
1. A system comprising:
a processor;
a camera;
a feature gallery for storing, for a plurality of objects, one or more multi-dimensional feature vectors extracted from images of the objects and one or more object identifiers associated with the objects; and
software executing on the processor, the software comprising;
a feature extractor for extracting feature vectors from a probe image collected using the camera;
a matching module for matching a product depicted in the probe image with a product stored in the feature gallery;
wherein the matching module calculates a distance between the one or more feature vectors extracted from the probe image and one or more feature vectors in the feature gallery by choosing, as a matching object, an object from the feature gallery having one or more feature vectors with the smallest distances from the one or more feature vectors extracted from the probe image; and
wherein the feature extractor is a multi-layered machine learning model comprising a plurality of trained convolutional neural networks, each convolutional neural network trained to extract a different type of feature vector characterizing a different feature from the probe image.
2. The system of claim 1 wherein the multi-layered machine learning model is trained on a dataset comprising object images and related object identifiers.
3. The system of claim 1 wherein the software further comprises:
an object identifier assignment module for assigning an object identifier to the object depicted in the probe image.
4. The system of claim 3 wherein the assigned object identifier is the object identifier associated with the object from the feature gallery matching the object depicted in the probe image.
5. The system of claim 1 further comprising:
a label association module for associating the object depicted in the probe image with a label.
6. The system of claim 5 wherein the label contains identifying information regarding the object.
7. The system of claim 5 :
wherein the feature gallery further comprises feature vectors extracted from a plurality of images of labels and associated object identifying information depicted on the label;
wherein the matching module can match an image of a label with a label stored in the feature gallery; and
wherein the matching module returns the object identifying information based on an input of a label image.
8. The system of claim 1 , further comprising:
a validation module for validating the assigned object identifier.
9. The system of claim 8 wherein the assigned object identifier is compared to the object identification information from a label associated with the object.
10. The system of claim 1 further comprising:
a stitching module for combining multiple images into a panoramic image.
11. The system of claim 10 wherein the stitching module combines overlapping images of adjacent areas by determining shared anchor points in each adjacent image and aligning the shared anchor points.
12. The system of claim 11 wherein the shared anchor points are objects in each image that share an object identifier.
13. The system of claim 11 wherein the shared anchor points are detected labels or label images having similar extracted feature vectors.
14. The system of claim 11 wherein the shared anchor points are a combination of objects in each image that share an object identifier and detected labels or label images having similar extracted feature vectors.