Parallel face detection and tracking system

• US 9,292,753 B2
• Filed: 11/21/2012
• Issued: 03/22/2016
• Est. Priority Date: 11/21/2012
• Status: Active Grant

• Current Assignee: Intel Corporation
• Original Assignee: Intel Corporation
• Inventors: Ranjan, Abhishek, Malik, Shahzad A., Hubert, Phillipp J., Patel, Umeshbhai, Javer, Faizal N., Mirza, Haroon F., Colson, William J.
• Primary Examiner(s): YANG, WEI WEN
• Application Number:
  US13/682,930
• Publication Number:
  US 20140140572A1
• Time in Patent Office: 1,217 Days
• Field of Search: None
• US Class Current: 1/1
• CPC Class Codes: G06V 10/96 Management of image or video recognition tasks
• CPC Class Codes: G06V 40/167 using comparisons between temporally consecutive images

Abstract

The present disclosure is directed to a parallel face detection and tracking system. In general, embodiments consistent with the present disclosure may be configured to distribute the processing load associated with the detection and tracking of different faces in an image between multiple data processors. If needed, processing load balancing and/or protective features may be implemented to prevent the data processors from becoming overwhelmed. In one embodiment, a device may comprise, for example, a communication module and at least one processing module. The communication module may be configured to receive at least image information that may be processed by a plurality of data processors in the data processing module. For example, each of the data processors may be configured to detect faces in the image information and/or track detected faces in the image information based on at least one criterion.

Claims

1. A device, comprising:

a communication module configured to receive image information;

at least one processing module including a plurality of data processors, each of the plurality of data processors being configured to perform at least one of a face detection workload or face tracking workload on the image information based at least in part on a plurality of face size scales; and

a face detection and tracking allocation module configured to;

determine said plurality of face size scales, wherein each of said plurality of face size scales corresponds at least in part to a size of at least one detected face in a plurality of detected faces in said image information, relative to a size of another detected face of said plurality of detected faces; and

balance performance of said face detection workload and said face tracking workload among the plurality of data processors at least in part by assigning different data processors of said plurality of data processors to different face size scales.

2. The device of claim 1, wherein the image information is a frame of video received from an image capture device.

3. The device of claim 2, further comprising a memory configured to store results information generated by each of the plurality of data processors, the results information including at least a detected face size and the location of each detected face within the frame.

4. The device of claim 2, wherein each of the plurality of data processors is further configured to delay performance of said face detection workload to a later received frame if performance of said face detection workload on a received frame would cause a processing load of a data processor to exceed a threshold.

5. The device of claim 1, wherein said face detection and tracking module is further to:

allocate performance of said face detection workload or said face tracking workload associated with relatively more process intensive face size scales to at least one first data processor of said plurality of data processors, and to allocate performance of said face detection workload or said face tracking workload associated with relatively less process intensive face size scales to at least one second data processor of said plurality of data processors; and

balance a face detection and tracking workload between the plurality of data processors such that relatively fewer face scales are allocated to said at least one first data processor, and relatively more face size scales are allocated to said at least one second data processor.

6. The device of claim 1, wherein the face detection and tracking allocation module is further to dynamically reallocate performance of said face detection workload and said face tracking workload among said plurality of data processors based at least one part on said one or more face size scales.

7. The device of claim 1, wherein the processing module is a microprocessor and the plurality of data processors are processing cores within the microprocessor.

8. A method, comprising:

determining a plurality of face size scales, wherein each face size scale corresponds at least in part to a size of one or more detected face in a plurality of detected faces in image information, relative to a size of another detected face of said plurality of detected faces;

determining how to allocate performance of at least one of face detection workload or face tracking workload on said image information among a plurality of data processors based at least in part on said plurality of face size scales; and

based on said determination, allocating performance of at least one of said face detection workload and said face tracking workload among the plurality of data processors at least in part by assigning different data processors of said plurality of processors to different face size scales.

9. The method of claim 8, wherein said determining how to allocate comprises determining said plurality of face size scales based at least in part on a maximum face size scale, a minimum face size scale, and a face size scale increment.

10. The method of claim 9, wherein:

said allocating comprises assigning performance of said face detection workload or said face tracking workload associated with a distribution of said plurality of face size scales to each of said plurality of data processors; and

said distribution is a linear distribution, such that each of the plurality of data processors receives an equal sized portion of said plurality of face size scales.

11. The method of claim 9, wherein:

said allocating comprises assigning performance of said face detection workload or said face tracking workload associated with a distribution of said plurality of face size scales to each of said plurality of data processors; and

said distribution is configured to balance a processing load among the plurality of data processors.

12. The method of claim 11, further comprising:

determining the processing load in each of the plurality of data processors; and

rebalancing the assignment of the performance of said face detection workload or said face tracking workload associated with said plurality of face size scales based on the determination of the processing load in each of the plurality of data processors.

13. The method of claim 8, wherein:

said allocating comprises assigning performance of said face detection workload or said face tracking workload associated with relatively more process intensive face scales to at least one first data processor of said plurality of data processors, and assigning performance of said face detection workload or said face tracking workload associated with relatively less process intensive face scales to at least one second data processor of said plurality of data processors; and

said allocating further comprises balancing at least one of said face detection workload and said face tracking workload among the plurality of data processors such that relatively fewer face scales are allocating to said at least one first data processor and relatively more face scales are allocating to said at least one second data processor.

14. The method of claim 8, further comprising:

assigning a threshold to at least one data processor of the plurality of data processors, the threshold being configured to cause the at least one data processor to delay performance of said face detection workload to a later received frame if performance of said face detection workload on a received frame would cause a processing load of in the at least one data processor to exceed the threshold.

15. At least one non-transitory machine-readable storage medium having stored thereon, individually or in combination, instructions that when executed by one or more processors result in the following operations comprising:

determining a plurality of face size scales, wherein each face size scale corresponds at least in part to a size of at least one detected face of a plurality of detected faces in image information, relative to a size of another detected face of said plurality of detected faces;

determining how to allocate performance of at least one of face detection workload or face tracking workload on said image information among a plurality of data processors based at least in part on said plurality of face size scales; and

based on said determination, allocating performance of at least one of said face detection workload and said face tracking workload among the plurality of data processors at least in part by assigning different data processors of said plurality of data processors to different face size scales.

16. The medium of claim 15, wherein said determining how to allocate comprises determining said plurality of face size scales based at least in part on a maximum face size scale, a minimum face size scale, and a face size scale increment.

17. The medium of claim 16, wherein:

said allocating comprises assigning performance of said face detection workload or said face tracking workload associated with a distribution of said plurality of face size scales to each of said plurality of data processors; and

said distribution is a linear distribution, such that each of the plurality of data processors receives an equal sized portion of said plurality of face size scales.

18. The medium of claim 16, wherein:

said allocating comprises assigning performance of said face detection workload or said face tracking workload associated with a distribution of said plurality of face size scales to each of said plurality of data processors; and

said distribution is configured to balance a processing load among the plurality of data processors.

19. The medium of claim 18, further comprising instructions that when executed by one or more processors result in the following operations comprising:

determining the processing load in each of the plurality of data processors; and

rebalancing the assignment of said face detection workload or said face tracking workload associated with said plurality of face size scales based on the determination of the processing load in each of the plurality of data processors.

20. The medium of claim 15, wherein:

said allocating comprises assigning performance of said face detection workload or said face tracking workload associated with relatively more process intensive face scales to at least one first data processor of said plurality of data processors, and allocating performance of said face detection workload or said face tracking workload associated with relatively less process intensive face scales to at least one second data processor of said plurality of data processors; and

said allocating further comprises balancing at least one of said face detection workload and said face tracking workload among the plurality of data processors such that relatively fewer face scales are allocating to said at least one first data processor and relatively more face scales are allocating to said at least one second data processor.

21. The medium of claim 15, further comprising instructions that when executed by one or more processors result in the following operations comprising:

assigning a threshold to at least one data processor of the plurality of data processors, the threshold being configured to cause the at least one data processor to delay performance of said face detection workload to a later received frame if performance of said face detection workload on a received frame would cause a processing load of in the at least one data processor to exceed the threshold.