Search International and National Patent Collections

1. (WO2018022393) MODEL-BASED CLASSIFICATION OF AMBIGUOUS DEPTH IMAGE DATA

Pub. No.:    WO/2018/022393    International Application No.:    PCT/US2017/042959
Publication Date: Fri Feb 02 00:59:59 CET 2018 International Filing Date: Fri Jul 21 01:59:59 CEST 2017
IPC: G06K 9/00
Applicants: MICROSOFT TECHNOLOGY LICENSING, LLC
Inventors: SCHOENBERG, Michael John
BLEYER, Michael
MESSER, Christopher S.
DEMANDOLX, Denis
Title: MODEL-BASED CLASSIFICATION OF AMBIGUOUS DEPTH IMAGE DATA
Abstract:
Ambiguous portions of an image which have fewer photons of a reflected light signal detected than required to determine depth can be classified as being dark (i.e., reflecting too few photons to derive depth) and/or far (i.e., beyond a range of a camera) based at least in part on expected depth and reflectivity values. Expected depth and reflectivity values for the ambiguous portions of the image may be determined by analyzing a model of an environment created by previously obtained images and depth and reflectivity values. The expected depth and reflectivity values may be compared to calibrated values for a depth sensing system to classify the ambiguous portions of the image as either dark or far based on the actual photon count detected for the image.