Object segmentation based on detected object-specific visual cues

摘要

One or more images of a physical environment may be received, where the one or more images may include one or more objects. A type of surface feature predicted to be contained on a portion of one or more surfaces of a single object may be determined. Surface features of the type within regions of the one or more images may then be identified. The regions may then be associated to corresponding objects in the physical environment based on the identified surface features. Based at least in part on the regions associated to the corresponding objects, a virtual representation of the physical environment may be determined, the representation including at least one distinct object segmented from a remaining portion of the physical environment so as to virtually distinguish a boundary of the at least one distinct object from boundaries of objects present in the remaining portion of the physical environment.

主权项

1. A method performed by a computing device having one or more processors and memory, the method comprising:
receiving one or more images of a physical environment, wherein the one or more images include one or more objects; based on the one or more images, determining one or more regions of the one or more images, the one or more regions representing estimated boundaries of distinct objects in the physical environment; determining a type of surface feature predicted to be contained on a portion of one or more surfaces of a single object; identifying, by the one or more processors, surface features of the type within the one or more regions; determining, for each region of the determined one or more regions, a respective confidence score based on whether an identified surface feature is contained in the region, wherein the respective confidence score is indicative of a probability that the region corresponds to a distinct object in the physical environment; and based on the determined confidence scores, determining, by the one or more processors, a virtual representation of the physical environment including at least one distinct object segmented from a remaining portion of the physical environment so as to virtually distinguish a boundary of the at least one distinct object from boundaries of objects present in the remaining portion of the physical environment.