Cloud based video detection and tracking system
A method for detecting and tracking multiple moving targets from airborne video within the framework of a cloud computing infrastructure. The invention simultaneously utilizes information from an optical flow generator and an active-learning histogram matcher in a complimentary manner so as to rule out erroneous data that may otherwise, separately, yield false target information. The invention utilizes user-based voice-to-text color feature description for track matching with hue features from image pixels.
1. A method for video detection and tracking of a target, comprising the steps of:
defining said target, wherein in said step of defining further comprises the steps of:
selecting a dataset of target image frames from a database; and
selecting the desired color of said target;
converting said color to a template hue histogram representation;
initializing an image detector;
performing target image frame alignment and registration in which homography matrices are generated;
generating an optical flow field;
performing morphology processing on said optical flow field so as to produce candidate target contours;
matching said target contours to said template hue histogram representation;
initializing tracking of said target and generating target tracks;
aligning current said target image frame so as to form a sequential track of said target; and
when said target is not located, redefining said target.
2. The method of claim 1 , wherein said step of generating an optical flow field further comprises generating an optical flow field between two aligned consecutive target image frames.
3. The method of claim 2 , further comprising the step of identifying strong target candidates from the occurrence of optical flow blobs in said optical flow field.
4. The method of claim 3 , further comprising the step of tuning said template hue histogram according to said strong target candidates.
5. The method of claim 4 , wherein said step of tuning said template hue histogram further comprises the steps of:
adapting color processing to the color characteristic of the light sourced under which said target image frames are captured; and
updating said template hue histogram in HSV color space to represent the color characteristics of said target image frame being captured.
6. The method of claim 5 , wherein said step of updating further comprises the steps of:
performing a weighted linear addition between said template hue histogram and a hue histogram corresponding to said strong target candidate; and
collecting color pixels of said strong target candidate.
7. The method of claim 1 , wherein said step of performing current target image frame alignment further comprises the steps of:
generating rotation and translation matrixes from said homography matrices; and
projecting, using said rotation and translation matrices, prior target frames of said target into a current target image frame.
8. The method of claim 6 , wherein the color depth of said hue histogram is divided into 16 segments over the value range of 0 to 255.
9. The method of claim 8 , wherein said step of weighted linear addition further comprises the steps of:
computing a current template histogram according to the expression:
C Template =α*C Template +β*C Candidate
wherein
C Template is the current template histogram;
α is a first weighting factor;
β is a second weighting factor; and
C Candidate is a strong candidate for a histogram window.
10. The method of claim 9 , wherein
α is selected as 0.9; and
β is selected as 0.1.
11. The method of claim 10 , further comprising the step of selecting candidate histogram windows with the most similar hue histogram as a true target.
12. The method of claim 7 , wherein said step of performing target image frame alignment and registration is performed in a cloud computing environment.
13. The method of claim 12 , and wherein said cloud computing environment is accessed through a Graphical User Interface (GUI).
14. The method of claim 13 , wherein said cloud computing environment is further accessed through voice command.
15. The method of claim 7 , wherein said homography matrix is a 3 by 3 matrix represented by
H
=
[
a
b
c
d
e
f
g
h
1
]
.
16. The method of claim 7 , wherein said rotation matrix R is represented by
[
a
b
d
e
]
said transformation matrix t is represented by
[
c
f
]
and projective parameters are represented by
[g h].
17. The method of claim 7 , further comprising the step of computing the coordinates of a track generated in a current frame, X new and y new , according to
x
new
=
a
x
+
b
y
+
c
g
x
+
h
y
+
1
and
y
new
=
d
x
+
e
y
+
f
g
x
+
h
y
+
1
.