Model-based feature tracking in 3-D and 2-D imagery
View Patent ↗A feature tracker for tracking a target includes an imaging sensor for imaging the target and a Kalman filter for generating predicted position, velocity and acceleration of the imaging sensor with respect to the target. The Kalman filter includes a state vector estimate of the position, velocity and acceleration of the imaging sensor, and a model for characterizing the target. The model characterizes the target by using at least one bivariate Gaussian function for the target. The Kalman filter includes a Jacobian matrix defined as a partial derivative of the model with respect to the state vector estimate. The Kalman filter includes a gain matrix generated from the Jacobian matrix.
1. A feature tracker for tracking features in a scene comprising:
an imaging sensor for imaging the target,
a Kalman filter for generating predicted position, velocity and acceleration of the imaging sensor with respect to the target,
the Kalman filter including
a state vector estimate including position, velocity and acceleration of the imaging sensor, and
a model for characterizing the target,
wherein the model includes at least one bivariate Gaussian function to model the target feature, and
the model includes at least two bivariate Gaussian functions, each bivariate Gaussian function defining a separate target in a terrain map.
2. The tracker of claim 1 wherein
the Kalman filter includes a Jacobian matrix defined as a partial derivative of the model with respect to the state vector estimate.
3. The tracker of claim 1 wherein
the measurement model includes a projection of the bivariate Gaussian function onto an image plane, and
the bivariate Gaussian function includes the form of
p
(
x
,
y
)
=
A
*
ⅇ
(
-
(
x
-
m
x
)
2
2
σ
x
2
-
(
y
-
m
y
)
2
2
σ
y
2
)
where A is the magnitude of the function,
m x and m y are centers of the function in an x, y plane, and
σ x and σ y define the width of the model in the x and y directions.
4. The tracker of claim 1 wherein
the Kalman filter includes a gain matrix generated from the Jacobian matrix , and
the gain matrix is multiplied by a difference between an actual measurement vector and an expected measurement vector to provide the state vector estimate.
5. The tracker of claim 1 wherein
the model includes a projection of a primitive function onto a plane of an image provided by the imaging sensor.
6. The tracker of claim 1 wherein
the model includes a primitive function defining the target.
7. The tracker of claim 1 wherein
the imaging sensor includes a LADAR, FLIR or camera, and
the model includes a terrain map with a primitive function defining the target, and
the primitive function is a bivariate Gaussian function.
8. The tracker of claim 1 wherein the model includes a Gaussian ellipse defining a three-dimensional structure projected onto a two-dimensional plane of an image taken by the imaging sensor.
9. A model of a terrain for use by a navigation/tracking system comprising:
a terrain map including at least one feature stored in a memory, and at least one primitive representing the feature stored in the memory,
wherein at least one primitive is substituted for the feature stored in the memory, and
at least two primitives represent two different features in the terrain map, and
the navigation/tracking system images the two different features and determines position, velocity and acceleration of the navigation/tracking system with respect to the two different features in the terrain map.
10. The model of a terrain in claim 9 wherein
the primitive is a bivariate Gaussian function.
11. The model of a terrain in claim 10 wherein
the bivariate Gaussian function has the form of
p
(
x
,
y
)
=
A
*
ⅇ
(
-
(
x
-
m
x
)
2
2
σ
x
2
-
(
y
-
m
y
)
2
2
σ
y
2
)
where A is the magnitude of the function,
m x and m y are centers of the function in an x, y plane, and
σ x and σ y define the width of the model in the x and y directions.
12. The model of a terrain in claim 10 including a LADAR, FLIR, or camera for providing a flight path for the navigation/tracking system.
13. A method of tracking a target comprising the steps of:
imaging the target by an imaging system;
predicting position, velocity and acceleration of the imaging system with respect to the target, using a Kalman filter;
modeling the target in the Kalman filter using a bivariate Gaussian function; and
generating a gain matrix in the Kalman filter by determining sensitivity of the bivariate Gaussian function to predicted position, velocity and acceleration of the imaging system, and
wherein the target includes at least two features in a terrain map, and
modeling includes replacing the two features with the two bivariate Gaussian functions.
14. The method of claim 13 wherein
the bivariate Gaussian function has the following form
p
(
x
,
y
)
=
A
*
ⅇ
(
-
(
x
-
m
x
)
2
2
σ
x
2
-
(
y
-
m
y
)
2
2
σ
y
2
)
wherein A is the magnitude of the function,
m x and m y are centers of the function in an x, y plane, and
σ x and σ y define the width of the model in the x and y directions.
15. The method of claim 14 including Θ,
where Θ defines a rotation of the model in the x,y plane.
16. The method of claim 13 including the step of:
generating a state vector including position, velocity and acceleration in the Kalman filter; and
wherein modeling the target includes generating a measurement matrix in the Kalman filter, and
determining the sensitivity of the bivariate Gaussian function includes calculating a partial derivative of the measurement matrix with respect to an estimate of the state vector.
17. The method of claim 13 wherein
imaging the target includes using one of either a LADAR, a FLIR, or a camera;
and
modeling the target includes replacing the target with the bivariate Gaussian function.