Particle tracking methods

1. A method of tracking an object in a video data, comprising:

determining a plurality of particles for use in estimating a location of the object in the video data;

allocating a first group of the plurality of particles to a graphics processing unit (GPU);

allocating a second group of the plurality of particles to one or more processing units;

determining a weight for each of the plurality of the particles, wherein:

the weight for at least some of the first and second groups of particles are determined in parallel;

the weight of two or more particles among the plurality of particles is determined substantially in parallel; and

the weight of each of the plurality of particles is based on a plurality of information cues and a Bhattacharyya similarity function; and

estimating the location of the object in the video data based upon the determined particle weights.

2. The method of claim 1 , comprising dividing the plurality of particles amongst the one or more processing units, wherein each processing unit is arranged to determine the weight for each of the particles allocated thereto.

3. The method of claim 1 , comprising dividing the plurality of particles amongst a plurality of processors in the GPU, wherein each processor is arranged to determine the weight for each of the particles allocated thereto.

4. The method of claim 1 , comprising selecting a particle having a greatest weight from the plurality of particles.

5. The method of claim 1 , wherein the plurality of information cues includes a colour histogram, an edge orientation histogram and wavelet features.

6. The method of claim 5 , wherein a weight P n colour of an nth particle based on the colour histogram is determined by:

P

n

colour

=

exp

⁢

{

-

D

2

⁡

(

H

n

colour

,

H

ref

colour

)

⁢

⁢

with

⁢

⁢

D

(

H

n

colour

,

H

ref

colour

⁢

)

=

(

1

-

∑

i

=

0

B

-

1

⁢

h

i

,

n

colour

·

h

i

,

ref

colour

)

1

/

2

wherein D is the Bhattacharyya similarity function, H n colour is a colour histogram in a HSV colour space for the nth particle and H ref colour is a reference colour histogram for the nth particle.

7. The method of claim 5 , wherein a weight P n edge of an nth particle based on the edge orientation histogram is determined by:

P

n

edge

=

exp

⁢

{

-

D

2

⁡

(

H

n

edge

,

H

ref

edge

)

⁢

⁢

with

⁢

⁢

D

⁡

(

H

n

edge

,

H

ref

edge

)

=

(

1

-

∑

i

=

0

B

-

1

⁢

h

i

,

n

edge

·

h

i

,

ref

edge

)

1

/

2

wherein H n edge is an edge orientation histogram for the nth particle, H ref edge is a reference edge orientation histogram and D is the Bhattacharyya similarity function.

8. The method of claim 1 , wherein the weight for each of the plurality of particles is determined as a sum of weights determined according to each of the plurality of information cues.

9. A non-transitory computer-useable medium having a compute readable program code to implement a method for tracking an object in video data comprising:

determining a plurality of particles for use in estimating a location of the object in the video data;

allocating a first group of the plurality of particles to a graphics processing unit (GPU);

allocating a second group of the plurality of particles to one or more processing units;

determining a weight for each of the plurality of the particles, wherein:

the weight for at least some of the first and second groups of particles are determined in parallel;

the weight of two or more particles among the plurality of particles is determined substantially in parallel; and

the weight of each of the plurality of particles is based on a plurality of information cues and a Bhattacharyya similarity function; and

estimating the location of the object in the video data based upon the determined particle weights.

10. An apparatus for tracking an object in a video data, comprising:

a first processing unit arranged to:

determine a plurality of particles; and

estimate a location of the object in the video data based upon a weight of each of the plurality of particles, wherein the weight of each of the plurality of particles is based on:

a plurality of information cues selected from a colour histogram, an edge orientation histogram and wavelet features; and

a Euclidean distance between a sample feature vector and a reference feature vector; and

a graphics processing unit arranged to determine a weight for at least some of the plurality of the particles, wherein the weight of two or more particles among the plurality of particles is determined substantially in parallel.

11. The apparatus of claim 10 , comprising a plurality of processing units, wherein the first processing unit is arranged to allocate a first group of the plurality of particles to the graphics processing unit and to allocate a second group of the plurality of particles to the plurality of processing units, wherein the graphics processing unit and the plurality of processing units are arranged to determine the weights for at least some of the first group and the second group in parallel.

12. The method of claim 10 , wherein a weight P n wavelet of an nth particle based on the edge orientation histogram is determined by:

P

n

wavelet

=

exp

⁢

{

-

Eu

⁡

(

V

n

wavelet

,

V

ref

wavelet

)

⁢

⁢

with

⁢

⁢

Eu

⁡

(

V

n

wavelet

,

V

ref

wavelet

)

=

(

∑

i

=

0

B

-

1

⁢

(

v

i

,

n

wavelet

-

v

i

,

ref

wavelet

)

2

)

1

/

2

wherein Eu is the Euclidean distance between the sample feature vector V n wavelet and the reference feature vector V ref wavelet .