System and method for identifying items

1. A system comprising non-transitory computer readable medium having stored thereon software instructions that, when executed by a processing system, cause the processing system to perform a method by executing steps, comprising:

capturing a measurement set of an item within a measurement volume, wherein the measurement set comprises imagery and a 3D measurement of the item;

determining a height map for the item based on the 3D measurement;

determining a region mask for the item based on the height map;

constructing a surface reconstruction for the item based on the region mask and, the 3D measurement; and

extracting a visual segment for the item from the imagery using the surface reconstruction, wherein the item is identified using the visual segment.

2. The system of claim 1 , further comprising:

a housing comprising a static base and defining the measurement volume; and

a camera statically mounted to the housing.

3. The system of claim 2 , wherein the camera comprises a stereo camera pair.

4. The system of claim 1 , wherein the steps further comprise: generating a point cloud from the measurement set, wherein the height map is generated using a top-down view of the point cloud.

5. The system of claim 1 , wherein the imagery comprises an image set, wherein each image of the image set is captured by a different camera from a different point of view.

6. The system of claim 5 , wherein each image of the image set is used to generate a different visual segment for the item, wherein the method further comprises:

extracting features from each visual segment; and

identifying the item based on the extracted features.

7. The system of claim 6 , wherein the features extracted from different visual segments are used to determine candidate identifiers, and wherein the item is identified based on the candidate identifiers using a majority voting algorithm.

8. The system of claim 1 , wherein identifying a region mask comprises detecting whether a different item is in contact with the item; and determining a different region mask for the different item when the item is in contact with the different item.

9. The system of claim 8 , wherein the region mask and the different region mask are determined using a segmentation module.

10. The system of claim 9 , wherein the segmentation module comprises a segmentation classifier.

11. The system of claim 1 , wherein the item comprises a transparent item, wherein identifying the region mask for the transparent item comprises identifying a set of continuous pixels associated with values that are outside of a predetermined range.

12. The system of claim 11 , wherein the values outside of the predetermined range comprise one of negative infinity, positive infinity, or no data.

13. The system of claim 1 , wherein the surface reconstruction comprises an upper geometry defined by the height map and vertical walls defined by projecting a boundary of the region mask along a vertical axis.

14. A system, comprising:

a sensor configured to sample a measurement set of an item comprising imagery and a 3D information of the item; and

a processing system, configured to:

determine a height map for the item based on the 3D information;

determine a region mask for the item based on the height map;

construct a surface reconstruction for the item using the region mask and, the 3D information; and

extract a visual segment for the item from the imagery using the surface reconstruction, wherein the item is identified using the visual segment.

15. The system of claim 14 , wherein the imagery comprises an RGB-Depth image.

16. The system of claim 15 , wherein the sensor comprises a camera associated with a field of view that encompasses a calibration pattern, wherein the processor is further configured to determine a calibration matrix for the camera based on the calibration pattern.

17. The system of claim 14 , wherein the visual segment is generated by projecting the surface reconstruction into the imagery.

18. The system of claim 14 , wherein the region mask is generated by feeding the height map as input into a segmentation classifier and receiving the region mask as output from the segmentation classifier.

19. The system of claim 18 , wherein the segmentation classifier is trained using a synthetic image, wherein the synthetic image comprises geometrically-combined item-specific geometric representation, wherein each item-specific geometric representation depicts a training item.

20. The system of claim 14 , wherein the surface reconstruction is a coarse mesh.

21. The system of claim 14 , wherein the surface reconstruction is generated by projecting the region mask downward along a z-axis of the height map.