System and method for the automatic enrollment of object images into a gallery
Disclosed herein is a system and method of identifying new 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 product library and a best-fit is obtained. A new product is identified if the distance between the features of the product on the shelf and the features of the best-fit product from the product library are above a predetermined threshold.
1. A computer-implemented method for automatically enrolling a new object in an object library comprising:
obtaining an image of a shelf having the object thereon, detecting a label in the image of the shelf;
extracting identifying information from the label;
determining a portion of the shelf associated with the label;
determining that the portion of the shelf associated with the label contains an object;
obtaining an image of the object;
determining a best-fit match between features extracted from the image of the object and features extracted from images of objects stored in the object library;
determining that a distance between the features extracted from the image of the object and features associated with the best-fit match falls above a predetermined threshold, indicating that the object in the object image is a new object;
adding the features extracted from the image of the object to the object library; and
adding the identifying information to the object library associated with the features extracted from the image of the object.
2. The method of claim 1 wherein the method is repeated for each label detected in the image of the shelf.
3. The method of claim 1 wherein the features are extracted from the image of the object by a trained feature extractor.
4. The method of claim 3 wherein the feature extractor outputs the identifier, given an image of the object as input.
5. The method of claim 1 wherein the object library is built by a method comprising:
obtaining a source image and identifying information for each object;
acquiring multiple images of each object from a plurality of sources;
ranking the acquired images based on a highest confidence in an association between the acquired images and the source image;
selecting a pre-determined number of the top-ranked acquired images; and
storing features extracted from the source image and the top-ranked acquired images, and the identifying information associated with the source image and the top-ranked acquired images, in the object library.
6. The method of claim 5 wherein the acquired multiple images for each object include images exhibiting different variations and/or viewpoints for each object.
7. The method of claim 6 wherein the feature extractor is trained on a dataset comprising multiple images of each object and an associated identifier.
8. The method of claim 6 wherein the multiple images of each object include images of the object exhibiting pose variations.
9. The method of claim 6 wherein the multiple images of each object include images of the object exhibiting variations in labelling of the object.
10. The method of claim 6 wherein the multiple images of each object include images of the object associated with different identifying information.
11. The method of claim 1 wherein different classes of objects in the object library have different predetermined thresholds.
12. A system for automatically enrolling a new object in an object library comprising:
a camera, for obtaining images of a shelf containing a plurality of objects and labels;
a processor, executing software;
a feature extractor trained to extract features from object images and output an identifier associated with an object in the object image; and
an object library containing features extracted from multiple images of various products, the feature associated with identifying information of the product;
wherein the software performs the functions of:
obtaining an image of a shelf having the object thereon,
detecting a label in the image of the shelf;
extracting identifying information from the label;
determining a portion of the shelf associated with the label;
determining that the portion of the shelf associated with the label contains an object;
obtaining an image of the object;
determining a best-fit match between features extracted from the image of the object and features extracted from images of objects stored in the object library;
determining that a distance between the features extracted from the image of the object and features associated with the best-fit match falls above a predetermined threshold, indicating that the object in the object image is a new object;
adding the features extracted from the image of the object to the object library; and
adding the identifying information to the object library associated with the features extracted from the image of the object.
13. The system of claim 12 wherein the method is repeated for each label detected in the image of the shelf.
14. The system of claim 12 wherein the feature extractor is a deep neural network trained on a dataset comprising multiple views of each object and associated identifiers.