Context-based priors for object detection in images

Patent metadata

Field	Value
Publication number	US-10410096-B2
Application number	US-201514882373-A
Country	US
Kind code	B2
Filing date	Oct 13, 2015
Priority date	Jul 9, 2015
Publication date	Sep 10, 2019
Grant date	Sep 10, 2019

Abstract

Context-based priors are utilized in machine learning networks (e.g., neural networks) for detecting objects in images. The likely locations of objects are estimated based on context labels. A machine learning network identifies a context label of an entire image. Based on the context label, the network selects a set of likely regions for detecting objects of interest in the image.

First claim

What is claimed is:

1. A method of object detection, comprising:
identifying, through a deep neural network (DNN), a context label corresponding to a scene of an entire image;
selecting a set of regions in the image expected to contain an object of interest based at least in part on the identified context label, the set of regions selected prior to locating the object of interest in the image; and
searching for the object of interest in the set of regions.

2. The method of claim 1 , further comprising training the DNN to refine the set of regions.

3. The method of claim 1 , further comprising creating the context label based at least in part on user input.

4. The method of claim 1 , further comprising creating the context label based at least in part on unsupervised learning.

5. The method of claim 1 , further comprising generating the set of regions based at least in part on the context label.

6. The method of claim 1 , further comprising:
identifying another context label; and
selecting another set of regions in the image expected to contain objects of interest based at least in part on the other identified context label.

7. The method of claim 1 , further comprising training the DNN to determine for each of the regions whether the object of interest is present.

8. The method of claim 1 , further comprising training the DNN to classify each of the regions according to the context label.

9. An apparatus for object detection, comprising:
a memory; and
at least one processor coupled to the memory, the at least one processor configured:
to identify, through a deep neural network (DNN), a context label corresponding to a scene of an entire image;
to select a set of regions in the image expected to contain an object of interest based at least in part on the identified context label, the set of regions selected prior to locating the object of interest in the image; and
search for the object of interest in the set of regions.

10. The apparatus of claim 9 , in which the at least one processor is further configured to train the DNN to refine the set of regions.

11. The apparatus of claim 9 , in which the at least one processor is further configured to create the context label based at least in part on user input.

12. The apparatus of claim 9 , in which the at least one processor is further configured to create the context label based at least in part on unsupervised learning.

13. The apparatus of claim 9 , in which the at least one processor is further configured to generate the set of regions based at least in part on the context label.

14. The apparatus of claim 9 , in which the at least one processor is further configured:
to identify another context label; and
to select another set of regions in the image expected to contain objects of interest based at least in part on the other identified context label.

15. The apparatus of claim 9 , in which the at least one processor is further configured to train the DNN to determine for each of the regions whether the object of interest is present.

16. The apparatus of claim 9 , in which the at least one processor is further configured to train the DNN to classify each of the regions according to the context label.

17. A non-transitory computer-readable medium for object detection having non-transitory program code recorded thereon, the program code comprising:
program code to identify, through a deep neural network (DNN), a context label corresponding to a scene of an entire image;
program code to select a set of regions in the image expected to contain an object of interest based at least in part on the identified context label, the set of regions selected prior to locating the object of interest in the image; and
searching for the object of interest in the set of regions.

18. The computer-readable medium of claim 17 , further comprising program code to train the DNN to refine the set of regions.

19. The computer-readable medium of claim 17 , further comprising program code to create the context label based at least in part on user input.

20. The computer-readable medium of claim 17 , further comprising program code to create the context label based at least in part on unsupervised learning.

21. The computer-readable medium of claim 17 , further comprising program code to generate the set of regions based at least in part on the context label.

22. The computer-readable medium of claim 17 , further comprising:
program code to identify another context label; and
program code to select another set of regions in the image expected to contain objects of interest based at least in part on the other identified context label.

23. The computer-readable medium of claim 17 , further comprising program code to train the DNN to determine for each of the regions whether the object of interest is present.

24. The computer-readable medium of claim 17 , further comprising program code to train the DNN to classify each of the regions according to the context label.

25. An apparatus for wireless communication, comprising:
means for identifying, through a deep neural network (DNN), a context label corresponding to a scene of an entire image;
means for selecting a set of regions in the image expected to contain an object of interest based at least in part on the identified context label, the set of regions selected prior to locating the object of interest in the image; and
means for searching for the object of interest in the set of regions.

26. The apparatus of claim 25 , further comprising means for training the DNN to refine the set of regions.

27. The apparatus of claim 25 , further comprising means for creating the context label based at least in part on unsupervised learning.

28. The apparatus of claim 25 , further comprising:
means for identifying another context label; and
means for selecting another set of regions in the image expected to contain objects of interest based at least in part on the other identified context label.

Assignees

• Qualcomm Inc

Inventors

• DIJKMAN Daniel Hendricus Franciscus
• Towal Regan Blythe
• Annapureddy Venkata Sreekanta Reddy

Classifications

• G06N3/02 Primary

  Neural networks · CPC title

• G06F18/24

  Classification techniques · CPC title

• G06V10/454

  Integrating the filters into a hierarchical structure, e.g. convolutional neural networks [CNN] · CPC title

• G06V10/768

  using context analysis, e.g. recognition aided by known co-occurring patterns · CPC title

• G06V10/255

  Detecting or recognising potential candidate objects based on visual cues, e.g. shapes · CPC title