Image Processing System and Method for Object Tracking

• US 20100177194A1
• Filed: 01/13/2010
• Published: 07/15/2010
• Est. Priority Date: 01/13/2009
• Status: Active Grant

• Current Assignee: Futurewei Technologies Incorporated (Huawei Investment & Holding Co., Ltd.)
• Original Assignee: Futurewei Technologies Incorporated (Huawei Investment & Holding Co., Ltd.)
• Inventors: Huang, Yu, Tian, Jun, Yu, Hong Heather, Li, Hongbing
• Granted Patent: US 8,345,102 B2
• Time in Patent Office: Days
• US Class Current: 348/157
• CPC Class Codes: G06T 2207/10024 Color image
• CPC Class Codes: G06T 2207/30221 Sports video; Sports image
• CPC Class Codes: G06T 7/277 involving stochastic approaches, e.g. using Kalman filters

Abstract

An image processing method can be performed on a video image that includes an initial frame and a plurality of subsequent frames. An object is located within the initial frame of the video image and a histogram related to the object is generated. A foreground map that includes the object is also generated. For each subsequent frame, a mean shift iteration is performed to adjust the location of the object within the current frame. The histogram related to the object and the foreground map can then be updated.

Claims

1. An image processing method comprising:

receiving a video image at an image processor, the video image comprising an initial frame and a plurality subsequent frames;

locating an object within an initial frame of the video image;

generating a histogram related to the object in the initial frame;

generating a foreground map that includes the object;

for each subsequent frame in the video image;

performing a mean shift iteration to adjust the location of the object within the current frame;

updating the histogram related to the object; and

updating the foreground map.

2. The method of claim 1, wherein performing the mean shift iteration comprises performing the mean shift iteration at a plurality of scales.

3. The method of claim 2, wherein updating the histogram related to the object comprises updating a scale.

4. The method of claim 2, wherein performing the mean shift iteration comprises:

performing a first search for the object within the current frame, the first search being performed with an original scale of the object;

performing a second search for the object within the current frame, the second search being performed with an enlarged scale of the object;

performing a third search for the object within the current frame, the third search being performed with a shrunken scale of the object; and

determining which search of the first, second and third searches found a best match of the object.

5. The method of claim 1, wherein performing the mean shift iteration comprises:

calculating a candidate normalized histogram;

processing the candidate normalized histogram with a mean shift vector; and

determining whether a result of the processing is less than a selected value.

6. The method of claim 1, wherein updating the histogram related to the object and updating the foreground map comprises:

determining that the object is lost;

relocating an object;

generating the histogram related to the object;

generating an updated foreground map that includes the object.

7. The method of claim 1, wherein locating the object comprises receiving information related to a location of where a user clicked on the video image.

8. The method of claim 1, wherein locating the object comprises executing an object location algorithm.

9. The method of claim 1, wherein locating the object comprises locating a player in a sports video.

10. An image processing method comprising:

receiving a video frame;

performing a first search for an object within the video frame by searching for a representation of the object within the video frame, the first search being performed with an original scale of the representation of the object;

performing a second search for the object within the video frame by searching for the representation of the object within the video frame, the second search being performed with an enlarged scale of the representation of the object;

performing a third search for the object within the video frame by searching for the representation of the object within the video frame, the third search being performed with a shrunken scale of the representation of the object; and

determining which search of the first, second and third searches found a best match of the object.

11. The method of claim 10, further comprising updating the representation of the object based on the determined best match.

12. The method of claim 11, wherein the representation comprises a size and a location.

13. The method of claim 10, wherein the representation of the object comprises a histogram.

14. The method of claim 10, wherein the enlarged scale is between about 5% and 15% larger than the original scale and the shrunken scale is between about 5% and 15% smaller than the original scale.

15. The method of claim 10, wherein, in each case, searching for the representation of the object comprises exhaustive searching.

16. The method of claim 10, wherein, in each case, searching for the representation of the object comprises a gradient descent searching based on a similarity measure.

17. An image processor comprising:

an input to receive a video image, the video image comprising an initial frame and a plurality of subsequent frames; and

a processing system to process the video image, the processing system including a processor programmed to perform the following steps;

locating an object within the initial frame of the video image;

generating a histogram related to the object in the initial frame;

generating a foreground map that includes the object;

for each subsequent frame in the video image;

performing a mean shift iteration to adjust the location of the object within the current frame;

updating the histogram related to the object; and

updating the foreground map.

18. The image processor of claim 17, wherein performing the mean shift iteration comprises:

calculating a candidate normalized histogram;

processing the candidate normalized histogram with a mean shift vector; and

determining whether a result of the processing is less than a selected value.

19. The image processor of claim 17, wherein updating the histogram related to the object and updating the foreground map comprises:

determining that the object is lost;

relocating an object;

generating the histogram related to the object;

generating an updated foreground map that includes the object.

20. The image processor of claim 17, wherein performing the mean shift iteration comprises:

performing a first search for the object within the current frame, the first search being performed with an original scale of the object;

performing a second search for the object within the current frame, the second search being performed with an enlarged scale of the object;

performing a third search for the object within the current frame, the third search being performed with a shrunken scale of the object; and

determining which search of the first, second and third searches found a best match of the object.