Occluded item detection for vision-based self-checkouts

1. A method, comprising:

training a first machine-learning algorithm on first images to identify first items based on non-occluded views of the first items present in the first images;

training a second machine-learning algorithm on second images to identify second items based on occluded views of pairs of the second items present in the second images;

receiving a transaction image of a transaction area during a transaction at a transaction terminal;

creating bounding boxes within the transaction image comprising first bounding boxes associated with the non-occluded views for the first items and second bounding boxes associated with the occluded views for the second items;

providing the first bounding boxes to the first machine-learning algorithm and receiving back first item identifiers for corresponding first items;

providing the second bounding boxes to the second machine-learning algorithm and receiving back second item identifiers for corresponding second items; and

processing the transaction with the first item identifiers and the second item identifiers.

2. The method of claim 1 wherein training the first machine-learning algorithm further includes providing the first images as multiple different camera angle views for each of the first items.

3. The method of claim 1 , wherein training the second machine-learning algorithm further includes providing the second images as multiple different occluded views for each pair of the second items.

4. The method of claim 1 , wherein receiving further includes cropping the transaction image as a cropped image by removing a known background scene present in the transaction image.

5. The method of claim 4 , wherein creating further includes identifying features from the cropped image and processing the features to identify the non-occluded views and the occluded views.

6. The method of claim 5 , wherein providing the first bounding boxes further includes providing the corresponding features associated with the first bounding boxes to the first machine-learning algorithm.

7. The method of claim 6 , wherein providing the second bounding boxes further includes providing the corresponding features associated with the second bounding boxes to the second machine-learning algorithm.

8. A system, comprising:

a camera;

a transaction terminal;

a server comprising a processor and a non-transitory computer-readable storage medium having executable instructions; and

the executable instructions when executed by the processor from the non-transitory computer-readable storage medium cause the processor to perform processing comprising:

obtaining a transaction image of a transaction area of the transaction from the camera;

cropping the transaction image as a cropped image by removing a known background scene associated with the transaction area;

extracting features from the cropped image;

identifying single items from the cropped image based on the features;

identifying multiple items that overlap and have an occluded view based on the features;

obtaining first item identifiers for the single items from a first trained-machine learning algorithm;

obtaining second item identifiers for the multiple items from a second trained-machine learning algorithm; and

processing the first item identifiers and the second item identifiers to complete a transaction at the transaction terminal.

9. The system of claim 8 , wherein the transaction terminal is a Self-Service Terminal (SST).

10. The system of claim 9 , wherein the SST is a vision-based SST that performs item recognition for transaction items based on the transaction image by providing the first item identifiers and the second item identifiers.