System and method for identifying items
In variants, a method for item recognition can include: optionally calibrating a sampling system, determining visual data using the sampling system, determining a point cloud, determining region masks based on the point cloud, generating a surface reconstruction for each item, generating image segments for each item based on the surface reconstruction, and determining a class identifier for each item using the respective image segments.
1 . A method for item classification, comprising:
determining a set of images of a set of items using a set of cameras;
determining a height map, separate from the set of images, for the set of items, wherein the height map comprises a top-down perspective of a measured volume;
determining a set of image segments, from the set of images, for each of the set of items based on the height map, using a segmentation model; and
determining an item label based on the set of image segments.
2 . The method of claim 1 , wherein the set of items is supported by a checkout surface physically connected to the set of cameras.
3 . The method of claim 1 , wherein the segmentation model is a segmentation classifier.
4 . The method of claim 1 , wherein the set of image segments are determined using instance-based segmentation.
5 . The method of claim 1 , wherein determining the item label comprises determining a feature vector from a set of intermediate layers of a neural network and determining the item label based on the feature vector.
6 . The method of claim 5 , wherein the neural network is trained using a training set of items that exclude the set of items.
7 . The method of claim 1 , wherein determining the item label comprises:
determining a label for each image segment of the set of image segments; and
determining the item label based on the labels for the image segments.
8 . The method of claim 1 , wherein the height map is determined based on measurements from a depth sensor with a different perspective from each of the set of cameras relative to the set of items.
9 . The method of claim 8 , wherein the height map and the set of images for the set of items are captured concurrently.
10 . The method of claim 1 , wherein determining the set of image segments comprises determining a volumetric segment using the height map and determining the set of image segments using the volumetric segment.
11 . The method of claim 10 , wherein determining the volumetric segment comprises determining a set of item boundaries from the height map and projecting the set of item boundaries vertically through the measured volume.
12 . The method of claim 10 , wherein determining the set of image segments using the volumetric segment comprises determining a surface reconstruction of the item using the volumetric segment and projecting the surface reconstruction into each image within the set of images.
13 . The method of claim 1 , further comprising: determining a trained model, wherein the item label is determined using the trained model.
14 . The method of claim 1 , wherein the set of items comprise items concurrently located within a measurement volume monitored by the set of cameras, wherein the method further comprises generating an invoice based on item labels for the set of items.
15 . The method of claim 1 , wherein the set of images are 2D images and the height map comprises more than 2 dimensions.
16 . The method of claim 11 , wherein the set of item boundaries are determined from the height map using a set of heuristics.
17 . The method of claim 1 , wherein determining the set of image segments comprise projecting a volumetric segment determined from the height map into each image within the set of images.
18 . A method for item classification, comprising:
determining a set of images of a set of items using a set of cameras;
determining a set of volumetric segments for the set of items;
segmenting the set of images into a set of image segments for the set of items, using the set of volumetric segments, comprising projecting each of the set of volumetric segments into each image within the set of images; and
determining an item label based on the set of image segments.
19 . The method of claim 18 , wherein the set of image segments are segmented after determining the set of volumetric segments.
20 . The method of claim 18 , wherein determining a set of volumetric segments comprises determining a set of item boundaries from a predetermined perspective of the measured volume, and projecting the set of item boundaries through the measured volume along a viewing axis of the predetermined perspective to define the set of volumetric segments.