VISUAL TRACKING SYSTEM AND METHOD THEREOF

• US 20090262977A1
• Filed: 09/03/2008
• Published: 10/22/2009
• Est. Priority Date: 04/18/2008
• Status: Active Grant

• Current Assignee: National Taiwan University (Government of The Republic of China)
• Original Assignee: National Taiwan University (Government of The Republic of China)
• Inventors: Fu, Li-Chen, Lin, Yi-Tzn, Hsiao, Pei-Yung, Huang, Cheng-Ming
• Granted Patent: US 8,885,876 B2
• Time in Patent Office: Days
• US Class Current: 382/103
• CPC Class Codes: G06T 2207/10016 Video; Image sequence
• CPC Class Codes: G06T 7/20 Analysis of motion motion estimation for coding, decoding, compressing or decompressing digital video signals H04N19/43, H04N19/51
• CPC Class Codes: H04N 23/61 based on recognised objects
• CPC Class Codes: H04N 23/695 Control of camera direction for changing a field of view, e.g. pan, tilt or based on tracking of objects

Abstract

The present invention provides a visual tracking system and its method comprising: a sensor unit, for capturing monitored scenes continuously; an image processor unit, for detecting when a target enters into a monitored scene, and extracting its characteristics to establish at least one model, and calculating the matching scores of the models; a hybrid tracking algorithm unit, for combining the matching scores to produce optimal matching results; a visual probability data association filter, for receiving the optimal matching results to eliminate the interference and output a tracking signal; an active moving platform, for driving the platform according to the tracking signal to situate the target at the center of the image. Therefore, the visual tracking system of the present invention can help a security camera system to record the target in details and maximize the visual information of the intruding target.

Claims

1. A visual tracking system, comprising:

a sensor unit, for capturing a monitored scene continuously;

an image processor unit, coupled to the sensor unit, for detecting when a target enters into the image, extracting characteristics from an image to formulate at least one model of the target, and calculating a matching score of the model;

a hybrid tracking algorithm unit, coupled to the image processor unit, for combining the matching score to produce an optimal matching result;

a visual probability data association filter, coupled to the hybrid tracking algorithm unit, for receiving the optimal matching result to remove an interference of tracking from a complicated background and output a tracking signal; and

an active moving platform, for driving a platform according to the tracking signal to situate the target at the center of an image.

2. The visual tracking system of claim 1, wherein at least two images are captured continuously.

3. The visual tracking system of claim 1, wherein at least one model can be a contour model, a color model or a gray scale intensity value model.

4. The visual tracking system of claim 1, wherein the image processor unit uses a binarization method and morphological method to determine when there is a target entering into the image.

5. The visual tracking system of claim 4, wherein the binarization method and morphological method are used to process a subtraction result of gray scale intensity values of two successive images, and the morphological method is used to obtain a connected component.

6. The visual tracking system of claim 5, when a threshold of pixel number of the connected component is greater than 400, then an indication of the target is shown.

7. The visual tracking system of claim 1, wherein the image processor unit creates a search area using the target position at a previous moment as a center, before at least one model of the target is created.

8. The visual tracking system of claim 7, wherein the model includes a template model and an active contour model.

9. The visual tracking system of claim 8, wherein the image processor unit is used to calculate a matching score of the template model by using an equation of the sum of absolute differences

10. The visual tracking system of claim 9, wherein the template size M×

N is 32×

32.

11. The visual tracking system of claim 8, wherein the image processor unit computes matching scores of the active contour model by an equation of gradient value summation

12. The visual tracking system of claim 11, wherein the number of sampling points is 40.

13. The visual tracking system of claim 1, wherein the hybrid tracking algorithm unit uses a fuzzy theory to integrate the matching scores to produce a matching result.

14. The visual tracking system of claim 13, wherein the matching result is a result selected from the previous ten results.

15. The visual tracking system of claim 1, wherein the visual probability data association filter is based on a Kalman filter.

16. A visual tracking method, comprising the steps of:

(a) using a sensor unit to continuously capture an image in a monitoring control scene, and determine when a target enters and obtain an initial location of the target;

(b) extracting a characteristic from the image to create at least one model of the target, and calculating the matching scores of the model;

(c) integrating the matching scores to produce a matching result;

(d) using the matching result to eliminate an interference of tracking a complicated background and output a tracking signal; and

(e) driving a platform to move according to the tracking signal to situate the target at the center of the image.

17. The visual tracking method of claim 16, wherein the step (a) further comprises the steps of:

(1) using the sensor unit to continuously capture at least two images in a monitored scene;

(2) using a binarization method and a morphological method to process the images to obtain a connected component;

(3) determining when the pixel number of the connected is greater than a threshold of the connected component, it indicates that a target enters into the monitored scene; and

(4) performing an intersection for a first binary image processed by the morphological method and an edge detection result of a second image to obtain an initial location of the target.

18. The visual tracking method of claim 17, wherein the step (2) further comprises the step of subtracting gray scale intensity values of two successive images to obtain a binary result, and using a morphological method to obtain a connected component.

19. The visual tracking method of claim 17, wherein the threshold of the connected component in the step (3) is 400.

20. The visual tracking method of claim 16, wherein the step (b) further comprises the steps of:

(1) using the target position of the previous moment as a center to create a search area from the continuously captured images;

(2) formulating a template model;

(3) initializing an active contour model;

(4) calculating a matching score of the template model; and

(5) calculating a matching score of the active contour model.

21. The visual tracking method of claim 20, wherein the matching score of the template model is calculated by using an equation of sum of absolute differences

22. The visual tracking method of claim 20, wherein the matching score of the active contour model is calculated by using an equation of gradient value summation

23. The visual tracking method of claim 16, wherein the step (c) uses a fuzzy theory to allocate the matching scores to produce a matching result.

24. The visual tracking method of claim 23, wherein the matching result is a result selected from the previous ten results.

25. The visual tracking method of claim 16, wherein the step (d) further comprises the steps of:

(1) using a Kalman filter to estimate possible position and speed of the target at the current moment from the target position of the previous moment in the optimal matching result;

(2) determining a possible range of the target position according to a covariance matrix and deleting the target position falling outside a validation gate;

(3) using an iterative threshold selection scheme to calculate a displacement between the target position falling within the validation gate and the target possible position falling within the validation gate, and deleting the target position if the displacement is smaller than a noise threshold; and

(4) outputting a tracking signal.