Method of tracking a human eye in a video image

• US 20070291983A1
• Filed: 06/14/2006
• Published: 12/20/2007
• Est. Priority Date: 06/14/2006
• Status: Active Grant

• Current Assignee: Delphi Technologies, Inc. (BorgWarner Incorporated)
• Original Assignee: Delphi Technologies, Inc. (BorgWarner Incorporated)
• Inventors: Hammoud, Riad I.
• Granted Patent: US 7,620,216 B2
• Time in Patent Office: Days
• US Class Current: 382/103
• 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: G06V 40/19 Sensors therefor

Abstract

A human eye is tracked between successively produced video image frames by consecutively applied eye location techniques. First, potential eye candidates within a local search window are identified using horizontal projection filtering, coupled with rudimentary appearance-based testing. The identified eye candidates are clustered and subjected to additional appearance-based testing to determine if one or more eye candidates should be accepted. If no eye candidates are accepted, a correlation/appearance technique is applied to the search window to identify the eye. If unsuccessful, the eye may be closed, and peripheral eye-related patches from the previous image frame are extracted and compared with the search window to identify the eye in the current frame. When comparable eye-related patches are identified in the search window for the current frame, the eye center is computed according to the mean of the eye periphery patches.

Claims

1. A method of tracking a target feature detected in a prior image frame to subsequent image frame, comprising the steps of:

defining a local search window in the subsequent image frame based on a location of the detected target in the prior image frame;

identifying target candidates from said local search window and accepting or rejecting the identified candidates based on appearance; and

if none of the identified candidates are accepted;

extracting a set of target-periphery patches from the prior image frame clustered about the location of the detected target;

searching the local search window to locate a set of image patches that correspond to the extracted set of target-periphery patches; and

if a corresponding set of image patches is located, determining a center location of the detected target in the subsequent image frame according to a center of the located set of image patches.

2. The method of claim 1, where said target feature is a human eye and the step of identifying target candidates from said local search window includes the steps of:

filtering and binarizing said local search window to form a target identification window;

projecting horizontals rows of pixels in said target identification window and selecting one or more rows of maximal receptive hole content;

identifying and clustering pixels disposed in the selected rows of maximal receptive hole content to form potential target candidates; and

testing said potential target candidates based on spatial and physiological considerations to select one or more winning target candidates.

3. The method of claim 2, including the steps of:

extracting target-sized patches of said subsequent image corresponding to the identified receptive hole pixels;

testing the extracted target-sized patches for target vs. non-target appearance and eliminating patches that exhibit non-target appearance; and

designating non-eliminated target-sized patches as target candidates.

4. The method of claim 1, where the extracted set of target-periphery patches are geometrically symmetrical with respect to the location of the detected target in the previous image frame.

5. The method of claim 1, where the step of searching the local search window includes the step of:

evaluating a similarity distance between the extracted target-periphery patches and various sets of potentially matching image patches in the local search window.

6. The method of claim 1, where said target feature is a human eye and the method includes the steps of:

extracting an image patch from the subsequent image frame centered on the determined location of the detected target; and

performing an appearance check on the extracted image patch to determine if it resembles a closed human eye.