Back to results
US 10,922,871 B2Granted
Casting a ray projection from a perspective view
AssigneeBAMTech, LLC
InventorsYuanyi Xue, Joseph Inzerillo, Dirk Van Dall
Life of this patent
Estimating the term…
Abstract
A device, system, and method cast a ray projection from a perspective view. The method includes determining a first mask for a first object from a first frame captured by a first camera. The method includes determining a second mask for the first object from a second frame captured by a second camera. The method includes generating a 3D mask by associating the first mask and the second mask. The method includes determining a location of the 3D mask. The method includes generating the ray projection of the 3D mask from a perspective of a second object.
Claims20
A method, comprising:
- identifying a first object in a first frame captured by a first camera using a semantic segmentation operation, the semantic segmentation operation including a branched training process that convolutes image data associated with the first object in the first frame along a plurality of branched paths having respective convolution parameters, the convoluted image data from each of the branched paths being combined into a combination convoluted data set;
- determining a first mask for the first object based on the first frame captured by the first camera and the combination convoluted data set;
- determining a second mask for the first object based on a second frame captured by a second camera;
- generating a 3D mask by associating the first mask and the second mask;
- determining a location of the 3D mask;
- and
- generating a ray projection of the 3D mask from a perspective of a second object.
4 more elements
A processing server, comprising:
- a transceiver configured to receive a first frame captured by a first camera and a second frame captured by a second camera;
- and
- a processor configured to identify a first object in a first frame captured by a first camera using a semantic segmentation operation, the semantic segmentation operation including a branched training process that convolutes image data associated with the first object in the first frame along a plurality of branched paths having respective convolution parameters, the convoluted image data from each of the branched paths being combined into a combination convoluted data set, the processor configured to determine a first mask for the first object based on the first frame and the combination convoluted data set and determine a second mask for the first object based on the second frame, the processor configured to generate a 3D mask by associating the first mask and the second mask, the processor configured to determine a location of the 3D mask, the processor configured to generate a ray projection of the 3D mask from a perspective of a second object.
A method, comprising:
- receiving a first frame captured by a first camera;
- performing a sequential training process on the first frame to identify a first object in the first frame based on the first frame, the sequential training process including a plurality of sequential pool operation sets configured sequentially to generate respective sequential pool feature sets, each of the sequential pool operation sets including a plurality of convolution operation sets generating respective convolution feature sets and a sequential pooling operation generating the respective sequential pool feature sets;
- performing a branched training process on the first frame to identify the first object in the first frame based on an output of the sequential training process, the branched training process including a plurality of branched pool operation sets configured in parallel to generate respective branched pool feature sets, each of the branched pool operation sets including a plurality of convolution operation sets generating respective convolution feature sets and a branched pooling operation generating the respective branched pool feature sets, the branched training process further including a combination operation to combine the branched pool feature sets into a combination data set;
- determining an accuracy estimate of a semantic boundary of the first object;
- and
- generating a first mask for the first object in the first frame based on the accuracy estimate.
3 more elements
Dependencies are inferred from the claim text.