Object analysis

Patent metadata

Field	Value
Publication number	US-11887366-B2
Application number	US-202017431002-A
Country	US
Kind code	B2
Filing date	Feb 11, 2020
Priority date	Feb 13, 2019
Publication date	Jan 30, 2024
Grant date	Jan 30, 2024

Abstract

A method comprising performing object detection within a set of representations of a hierarchically-structured signal, the set of representations comprising at least a first representation of the signal at a first level of quality and a second representation of the signal at a second, higher level of quality.

First claim

The invention claimed is:

1. A method of performing object detection within a set of representations of a hierarchically-structured signal, wherein the hierarchically-structured signal is structured in accordance with a tiered hierarchy of representations of an original signal and each of the representations is associated with respective level of quality, the set of representations comprising at least a first representation of the original signal at a first level of quality and a second representation of the original signal at a second, higher level of quality, wherein the second representation is based on reconstruction data used to adjust the first representation, the method comprising:
performing object detection within one or more representations to find one or more instances of one or more objects of one or more particular classes and localizing the one or more objects within the one or more representations.

2. A method according to claim 1 , wherein said object detection is performed using at least one convolutional neural network (CNN).

3. A method according to claim 2 , wherein said object detection is performed using a first CNN associated with the first level of quality and a second CNN associated with the second level of quality.

4. A method according to claim 3 , wherein data output by the first CNN is provided to the second CNN.

5. A method according to claim 1 , comprising performing object recognition within one or more of the set of representations.

6. A method according to claim 1 , comprising obtaining the first representation by decoding a tier of the hierarchically-structured signal, wherein the hierarchically-structured signal is encoded in a hierarchically-encoded signal received from an encoder.

7. A method according to claim 1 , comprising obtaining at least part of the second representation using at least part of the first representation.

8. A method according to claim 7 , wherein the at least part of the second representation is obtained in response to determining that a level of confidence associated with object recognition performed within the first representation does not meet an object-recognition threshold level of confidence.

9. A method according to claim 7 , comprising obtaining only part of the second representation using only part of the first representation.

10. A method according to claim 9 , wherein object detection and/or object recognition is performed within the part of the second representation.

11. A method according to claim 10 , wherein the part of the first representation corresponds to a region of interest within the first representation.

12. A method according to claim 1 , wherein the signal comprises a video signal, wherein the first and second representations are each of the same time sample of the video signal, wherein the level of quality corresponds to an image resolution.

13. A method according to claim 1 , wherein the set of representations comprises a third representation of the signal at a third level of quality, the third level of quality being higher than the second level of quality.

14. A method according to claim 1 , wherein the method is performed in a hierarchical system.

15. An apparatus configured to perform object detection detection within a set of representations of a hierarchically-structured signal, wherein the hierarchically-structured signal is structured in accordance with a tiered hierarchy of representations of an original signal and each of the representations is associated with respective level of quality, the set of representations comprising at least a first representation of the original signal at a first level of quality and a second representation of the original signal at a second, higher level of quality, wherein the second representation is based on reconstruction data used to adjust the first representation, the apparatus comprising:
a processor;
a non-tranitory computer-readable medium having stored therein computer executable instructions that, when executed by the processor, cause the apparatus to:
perform object detection within one or more representations to find one or more instances of one or more objects of one or more particular classes and localizing the one or more objects within the one or more representations.

16. A non-transitory computer-readable medium having stored therein computer executable instructions that, when executed by a processor of a computing system, cause the computing system to perform object detection within a set of representations of a hierarchically-structured signal, wherein the hierarchically-structured signal is structured in accordance with a tiered hierarchy of representations of an original signal and each of the representations is associated with respective level of quality, the set of representations comprising at least a first representation of the original signal at a first level of quality and a second representation of the original signal at a second, higher level of quality, wherein the second representation is based on reconstruction data used to adjust the first representation, the computing system caused to:
perform object detection within one or more representations to find one or more instances of one or more objects of one or more particular classes and localizing the one or more objects within the one or more representations.

Assignees

• V Nova Int Ltd

Inventors

• Meardi Guido
• Cobianchi Guendalina
• Keszthelyi Balázs
• Makeev Ivan
• Ferrara Simone
• Poularakis Stergios

Classifications

• G06N3/0455

  Auto-encoder networks; Encoder-decoder networks · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06V10/774

  Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title

• G06V10/764

  using classification, e.g. of video objects · CPC title

• G06V10/82

  using neural networks · CPC title