MODEL COMPRESSION IN BINARY CODED IMAGE BASED OBJECT DETECTION

• US 20160171344A1
• Filed: 12/11/2014
• Published: 06/16/2016
• Est. Priority Date: 12/11/2014
• Status: Active Grant

• Current Assignee: Intel Corporation
• Original Assignee: Intel Corporation
• Inventors: CHEN, KE, JUN, BONGJIN, KIM, DUKHWAN, CHIU, YI-JEN, KIM, TAE-HOON
• Granted Patent: US 9,697,443 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06F 18/2148 characterised by the process organisation or structure, e.g. boosting cascade
• CPC Class Codes: G06F 18/24 Classification techniques
• CPC Class Codes: G06V 10/443 by matching or filtering
• CPC Class Codes: G06V 10/467 Encoded features or binary features, e.g. local binary patterns [LBP]
• CPC Class Codes: G06V 10/7515 Shifting the patterns to accommodate for positional errors
• CPC Class Codes: G06V 10/7747 Organisation of the process, e.g. bagging or boosting
• CPC Class Codes: G06V 40/161 Detection; Localisation; Normalisation
• CPC Class Codes: H04N 19/90 using coding techniques not provided for in groups H04N19/10-H04N19/85, e.g. fractals

Abstract

Techniques related to object detection using binary coded images are discussed. Such techniques may include performing object detection based on multiple spatial correlation mappings between a generated binary coded image and a binary coded image based object detection model and nesting look up tables such that binary coded representations are grouped and such groups are associated with confidence values for performing object detection.

Claims

1. A computer-implemented method for performing object detection comprising:

generating a binary coded image based on individual pixels of at least a portion of an input image, the binary coded image comprising a plurality of binary coded representations each associated with a pixel of the individual pixels;

performing a first object detection based on a first spatial correlation mapping between the binary coded image and a binary coded image based object detection model; and

performing a second object detection based on a second spatial correlation mapping between the binary coded image and the binary coded image based object detection model.

2. The method of claim 1, wherein the first spatial correlation comprises first pixel coordinates for a look up table of the binary coded image based object detection model indicating a first position within the portion of the input image and the second spatial correlation comprises second pixel coordinates for the look up table of the binary coded image based object detection model indicating a second position within the portion of the input image.

3. The method of claim 1, wherein the first object detection is based at least in part on accessing a look up table for a first binary coded representation of a first pixel location within the portion of the input image and the second object detection is based at least in part on accessing the look up table for a bit shifted binary coded representation of a second binary coded representation of a second pixel location within the portion of the input image.

4. The method of claim 3, wherein the second pixel location is a 90 degree rotation from the first pixel location and the bit shifted binary coded representation comprises a circular shift by two bits of the second binary coded representation of the second pixel location.

5. The method of claim 3, wherein the bit shifted binary coded representation comprises at least two bit shifting operations on the second binary coded representation of the second pixel location.

6. The method of claim 1, wherein the binary coded image based object detection model represents at least one of a vertically aligned object or a 30 degree offset object, and wherein the first spatial correlation and the second spatial correlation comprise a spatial relationship comprising at least one of a 90 degree clockwise rotation, a 90 degree counterclockwise rotation, or a mirror relationship.

7. The method of claim 1, wherein the binary coded image based object detection model comprises, for a first pixel of the individual pixels, two nested look up tables comprising a first look up table comprising group numbers indexed by binary coded representation values and a second look up table comprising confidence values indexed by the group numbers.

8. The method of claim 1, further comprising:

performing a third object detection based on a third spatial correlation mapping between the binary coded image and a second binary coded image based object detection model; and

performing a fourth object detection based on a fourth spatial correlation mapping between the binary coded image and the second binary coded image based object detection model.

9. The method of claim 1, wherein generating the binary coded image comprises generating a first binary coded representation of the plurality of binary coded representations for a first pixel of the individual pixels by assigning, to each of a first plurality of neighbor pixels of the first pixel, a value of 0 or 1 based on a comparison of each pixel value of the first plurality of neighbor pixels to a threshold for the first individual pixel and determining the first binary coded representation based on a sequencing of the assigned values of the first plurality of neighbor pixels.

10. A computer-implemented method for performing object detection comprising:

accessing a first look up table comprising group numbers indexed by binary coded representation values to determine an individual group number associated with an individual binary coded representation of a binary coded image representing an input image;

accessing, based at least in part on the individual group number, a second look up table comprising confidence values to determine an individual confidence value associated with the individual binary coded representation; and

performing an object detection for the input image based at least in part on the individual confidence value.

11. The method of claim 10, wherein the second look up table is indexed based on the group numbers.

12. The method of claim 10, wherein the second look up table is indexed based on second group numbers, the method further comprising:

accessing a third look up table comprising the second group numbers indexed by the group numbers to determine an individual second group number associated with the individual binary coded representation.

13. The method of claim 10, wherein the first look up table is associated with pixel coordinates of a location within a scan window of the input image, the method further comprising:

adjusting the pixel coordinates associated with the first look up table to provide an adjusted spatial correlation to the binary coded image.

14. The method of claim 13, further comprising:

determining a second binary coded representation for a second location within the scan window associated with the adjusted coordinates;

shifting bits of the second binary coded representation to generate a bit shifted second binary coded representation;

accessing the first look up table to determine a second individual group number associated with the bit shifted second binary coded representation;

accessing, based at least in part on the second individual group number, the second look up table to determine a second individual confidence value associated with the second individual binary coded representation; and

performing a second object detection for the input image based at least in part on the second individual confidence value.

15. A system for providing object detection on a computing device, comprising:

a memory configured to store a binary coded image based object detection model; and

a processing unit coupled to the display device, wherein the processing unit comprises;

binary coded image circuitry to generate a binary coded image based on individual pixels of at least a portion of an input image, the binary coded image comprising a plurality of binary coded representations each associated with a pixel of the individual pixels; and

object detection circuitry to perform a first object detection based on a first spatial correlation mapping between the binary coded image and the binary coded image based object detection model and to perform a second object detection based on a second spatial correlation mapping between the binary coded image and the binary coded image based object detection model.

16. The system of claim 15, wherein the first spatial correlation comprises first pixel coordinates for a look up table of the binary coded image based object detection model associated with a first position within the portion of the input image and the second spatial correlation comprises second pixel coordinates for the look up table of the binary coded image based object detection model associated with a second position within the portion of the input image.

17. The system of claim 15, wherein the first object detection is based at least in part on a first look up table access for a first binary coded representation of a first pixel location within the portion of the input image and the second object detection is based at least in part on a second access of the look up table for a bit shifted binary coded representation of a second binary coded representation of a second pixel location within the portion of the input image.

18. The system of claim 15, wherein the binary coded image based object detection model comprises, for a first pixel of the individual pixels, two nested look up tables comprising a first look up table comprising group numbers indexed by binary coded representation values and a second look up table comprising confidence values indexed by the group numbers.

19. The system of claim 15, wherein the object detection circuitry is to perform a third object detection based on a third spatial correlation mapping between the binary coded image and a second binary coded image based object detection model and perform a fourth object detection based on a fourth spatial correlation mapping between the binary coded image and the second binary coded image based object detection model.

20. A system for providing object detection on a computing device, comprising:

a memory configured to store a first look up table and a second look up table; and

a processing unit coupled to the display device, wherein the processing unit comprises;

object detection circuitry to access the first look up table comprising group numbers indexed by binary coded representation values to determine an individual group number associated with an individual binary coded representation for a pixel of an input image, access, based at least in part on the individual group number, the second look up table comprising confidence values to determine an individual confidence value associated with the individual binary coded representation, and perform object detection for the input image based at least in part on the individual confidence value.

21. The system of claim 20, wherein the first look up table is associated with pixel coordinates of a location within a scan window of the input image, and wherein the object detection circuitry is further to adjust the pixel coordinates associated with the first look up table to provide an adjusted spatial correlation to the binary coded image.

22. The system of claim 21, wherein the object detection circuitry is further to receive a second binary coded representation for a second location within the scan window associated with the adjusted coordinates, shift bits of the second binary coded representation to generate a bit shifted second binary coded representation, access the first look up table to determine a second individual group number associated with the bit shifted second binary coded representation, access, based at least in part on the second individual group number, the second look up table to determine a second individual confidence value associated with the second individual binary coded representation, and perform a second object detection for the input image based at least in part on the second individual confidence value.

23. At least one machine readable medium comprising a plurality of instructions that in response to being executed on a computing device, cause the computing device to perform object detection by:

generating a binary coded image based on individual pixels of at least a portion of an input image, the binary coded image comprising a plurality of binary coded representations each associated with a pixel of the individual pixels;

performing a first object detection based on a first spatial correlation mapping between the binary coded image and a binary coded image based object detection model; and

performing a second object detection based on a second spatial correlation mapping between the binary coded image and the binary coded image based object detection model.

24. The machine readable medium of claim 23, wherein the first spatial correlation comprises first pixel coordinates for a look up table of the binary coded image based object detection model indicating a first position within the portion of the input image and the second spatial correlation comprises second pixel coordinates for the look up table of the binary coded image based object detection model indicating a second position within the portion of the input image.

25. The machine readable medium of claim 23, wherein the first object detection is based at least in part on accessing a look up table for a first binary coded representation of a first pixel location within the portion of the input image and the second object detection is based at least in part on accessing the look up table for a bit shifted binary coded representation of a second binary coded representation of a second pixel location within the portion of the input image.

26. The machine readable medium of claim 23, wherein the binary coded image based object detection model comprises, for a first pixel of the individual pixels, two nested look up tables comprising a first look up table comprising group numbers indexed by binary coded representation values and a second look up table comprising confidence values indexed by the group numbers.

27. The machine readable medium of claim 23, the machine readable medium comprising further instructions that in response to being executed on a computing device, cause the computing device to perform object detection by:

performing a third object detection based on a third spatial correlation mapping between the binary coded image and a second binary coded image based object detection model; and

performing a fourth object detection based on a fourth spatial correlation mapping between the binary coded image and the second binary coded image based object detection model.

28. At least one machine readable medium comprising a plurality of instructions that in response to being executed on a computing device, cause the computing device to perform object detection by:

accessing a first look up table comprising group numbers indexed by binary coded representation values to determine an individual group number associated with an individual binary coded representation of a binary coded image representing an input image;

accessing, based at least in part on the individual group number, a second look up table comprising confidence values to determine an individual confidence value associated with the individual binary coded representation; and

performing an object detection for the input image based at least in part on the individual confidence value.

29. The machine readable medium of claim 28, wherein the first look up table is associated with pixel coordinates of a location within a scan window of the input image, the machine readable medium comprising further instructions that in response to being executed on a computing device, cause the computing device to perform object detection by:

adjusting the pixel coordinates associated with the first look up table to provide an adjusted spatial correlation to the binary coded image.

30. The machine readable medium of claim 29, the machine readable medium comprising further instructions that in response to being executed on a computing device, cause the computing device to perform object detection by:

determining a second binary coded representation for a second location within the scan window associated with the adjusted coordinates;

shifting bits of the second binary coded representation to generate a bit shifted second binary coded representation;

accessing the first look up table to determine a second individual group number associated with the bit shifted second binary coded representation;

accessing, based at least in part on the second individual group number, the second look up table to determine a second individual confidence value associated with the second individual binary coded representation; and

performing a second object detection for the input image based at least in part on the second individual confidence value.