Loading patent…
AssigneeBoard of Trustees of Michigan State University
InventorsDaniel Morris, Steven Yik, Madonna Benjamin, Michael Lavagnino
Estimating the term…
This disclosure describes various attributes and implementations of systems and methods for efficiently generating high accuracy landmark annotations for depth-based image or video data sets. For example, a Transpositional Tagging approach can automatically or semi-automatically find, identify, and track landmarks (such as human or animal joints, other structural landmarks, or other points of interest) that are visible in one imaging modality (such as infrared, optical, etc.), and transfer those labels to a second image modality (such as, e.g., near-IR depth images/videos). As a result, systems and methods provided herein can quickly generate highly specific training datasets that are then used to develop neural network models for detecting poses, positions, and/or movements in imaging modalities such as 3D and depth images.
A system comprising:
A method comprising:
Dependencies are inferred from the claim text.