Depth-based object re-identification
AssigneeMicrosoft Technology Licensing, LLC
InventorsNikolaos Karianakis, Zicheng Liu, Yinpeng Chen
Life of this patent
Estimating the term…
Abstract
An object re-identifier. For each of a plurality of frames of a video, a quality of the frame is assessed and a confidence that a previously-recognized object is present in the frame is determined. The determined confidence for the frame is weighted based on the assessed quality of the frame such that frames with higher relative quality are weighted more heavily than frames with lower relative quality. An overall confidence that the previously-recognized object is present in the video is assessed based on the weighted determined confidences.
Claims20
A method, comprising:
- for each of a plurality of frames of a video captured by a single camera, assessing a quality of the frame;
- for each of the plurality of frames of the video captured by the single camera, determining a confidence that a previously-recognized object is present in the frame;
- for each of the plurality of frames of the video captured by the single camera, weighting the determined confidence for the frame based on the assessed quality of the frame such that frames with higher relative quality are weighted more heavily than frames with lower relative quality;
- and
- assessing an overall confidence that the previously-recognized object is present in the video captured by the single camera based on the weighted determined confidences.
A storage machine holding instructions executable by a logic machine, the instructions comprising:
- instructions to, for each of a plurality of frames of a video captured by a single camera, assess a quality of the frame;
- instructions to, for each of the plurality of frames of the video captured by the single camera, determine a confidence that a previously-recognized object is present in the frame;
- instructions to, for each of the plurality of frames of the video captured by the single camera, weight the determined confidence for the frame based on the assessed quality of the frame such that frames with higher relative quality are weighted more heavily than frames with lower relative quality;
- and
- instructions to, assess an overall confidence that the previously-recognized object is present in the video captured by the single camera based on the weighted determined confidences.
A computing system comprising:
- a logic machine;
- and
- a storage machine holding instructions executable by the logic machine to instantiate:
- a re-identification model previously-trained to receive data derived from a video including a plurality of frames, and output an overall confidence indicating that a previously-recognized object is present in the video, the re-identification model including:
- a frame-level model configured to, for each of the plurality of frames, receive the data derived from the frame, and output a frame-level confidence indicating that the previously-recognized object is present in the frame;
- a temporal attention model configured to, for each of the plurality of frames, receive the data derived from the frame, and output a weight based on a quality of the frame, wherein frames with higher relative quality are weighted more heavily than frames with lower relative quality;
- a recurrent model configured to, for each of the plurality of frames, output a current integrated confidence based on the frame-level confidence output by the frame-level model for the frame and a previous integrated confidence output by the recurrent model for a previous frame;
- and
- a classifier model configured to, for each of the plurality of frames, determine a weighted confidence for the frame based on the current integrated confidence output from the recurrent model and the weight output from the temporal attention model, and output the overall confidence based on the weighted confidences of the plurality of frames.
Dependencies are inferred from the claim text.