Image processing for occluded item recognition
Cameras capture time-stamped images of predefined areas. Individuals and items are tracked in the images. Occluded items detected in the images are preprocessed to remove pixels associated with occluded information and the pixels remaining associated with the items are cropped. The preprocessed and cropped images are provided to a trained machine-learning algorithm or a trained neural network trained to classify and identify the items. Output received from the trained neural network provides item identifiers for the items that are present in the original images.
1. A method, comprising:
receiving an occluded image associated with an unknown item that is in the possession of a tracked person within a store;
modifying pixels of the occluded image producing a modified image representing unknown item pixels distinguished from other pixels, wherein modifying further includes removing tracked person pixels that depict the tracked person from the modified image by clustering the pixels into buckets and removing a given bucket of pixels associated with the tracked person pixels, wherein modifying further includes segmenting the modified image to isolate the unknown item pixels based on detected edges and removing the tracked person pixels;
providing the modified image to a trained machine-learning algorithm as an input;
receiving as an output from the trained machine-learning algorithm a known item identifier for a known item associated with the unknown item; and
providing the known item identifier to a transaction manager that manages a shopping cart for a transaction for the tracked person within the store.
2. The method of claim 1 , wherein the modifying further includes identifying the edges of the unknown item within the occluded image.
3. The method of claim 2 , wherein the identifying of the edges further includes blacking out background pixels associated with an area within the occluded image that falls outside a second area defined within the edges.
4. The method of claim 3 , wherein the blacking out further includes cropping the background pixels within the modified image.
5. The method of claim 4 , wherein the cropping further includes blacking out portions of the second area associated with a hand or a limb of the tracked person within the modified image.
6. The method of claim 1 , wherein the providing the modified image further includes obtaining a list of potential items for the unknown item based on a location within the store where the occluded image was captured, and providing the list as additional input to the trained machine-learning algorithm.
7. The method of claim 1 , wherein the receiving the output further includes receiving a list of potential item identifiers as the output from the trained machine-learning algorithm and selecting the known item identifier from the list.
8. The method of claim 1 , wherein the providing of the known item identifier further includes providing a tracked person identifier for the tracked person to the transaction manager with the known item identifier.