Method and system for motion prediction
AssigneeIndustrial Technology Research Institute
InventorFu-Song HSU
Life of this patent
Estimating the term…
Abstract
A method for motion prediction is provided. The method includes: capturing, by a single camera, a plurality of cameras or 3D software, images of an object at multiple angles to generate multi-view images of the object; synthesizing motion capture (MoCap) data according to the multi-view images; projecting a masking object onto the object to generate multi-view training images, wherein the multi-view training images are images in which parts of limbs of the object are unoccluded and other parts of the limbs of the object are occluded; and using the motion capture data and the multi-view training images to train a predictive model.
Claims18
A method for motion prediction, comprising:
- capturing, by a single camera, a plurality of cameras or 3D software, images of an object at multiple angles to generate multi-view images of the object;
- synthesizing motion capture (MoCap) data according to the multi-view images;
- projecting a masking object onto the object to generate multi-view training images, wherein the multi-view training images are images in which parts of limbs of the object are unoccluded and other parts of the limbs of the object are occluded;
- and
- using the motion capture data and the multi-view training images to train a predictive model.
A system for motion prediction, comprising:
- a device for motion prediction, wherein the device for motion prediction executes:
- capturing, by a single camera, a plurality of cameras or 3D software, images of an object at multiple angles to generate multi-view images of the object;
- synthesizing motion capture (MoCap) data according to the multi-view images;
- projecting a masking object onto the object to generate multi-view training images, wherein the multi-view training images are images in which parts of limbs of the object are unoccluded and other parts of the limbs of the object are occluded;
- and
- using the motion capture data and the multi-view training images to train a predictive model.
Dependencies are inferred from the claim text.