Three dimensional position estimation mechanism

1. An apparatus to facilitate three dimensional (3D) position estimation, comprising:

one or more processors to receive a plurality of two dimensional (2D) images captured by a camera array during a live event, locate key-points of human joints of a plurality of event participants included in the images, associate key-points of each participant across the images and recover a 3D body position of each of the plurality of participants based on the associated key-points, wherein locating the key-points of the human joints comprises detecting major joints of each of the plurality of participants associated with a bounding box and labeling the key-points at the major joints, and

wherein the one or more processors are further to perform triangulation of a position of each of the major joints in a 3D space for a participant having the human joints, wherein the triangulation is performed based on camera projection matrices and pixel-locations of the major joints across the images.

2. The apparatus of claim 1 , wherein the one or more processors further generates an animated model including animated motion for one or more of the plurality of participants.

3. The apparatus of claim 1 , wherein associating the key-points of each participant across the images comprises determining a geometric pairwise correspondence between the images and determine one or more groups of images associated with a same participant from a plurality of camera angles.

4. The apparatus of claim 3 , wherein the one or more processors further defines a region of interest within the images, detects each of the plurality of participants in the region of interest, and generates a bounding box associated with each of the plurality of participants.

5. The apparatus of claim 1 , wherein the major joints are detected via a convolution neural network (CNN).

6. The apparatus of claim 1 , wherein the one or more processors further perform kinematic model fitting to capture a position and orientation of one or more major segments of each of the plurality of participants to generate a model corresponding to each of the plurality of participants.

7. The apparatus of claim 6 , wherein the one or more processors further perform temporal association to fit a model of each of the plurality of participants across a plurality of video frames.

8. A method to facilitate three dimensional (3D) position estimation, comprising:

receiving a plurality of two dimensional (2D) images captured by a camera array during a live event;

locating key-points of human joints of a plurality of event participants included in the images, wherein locating the key-points of the human joints comprises detecting major joints of each of the plurality of participants associated with a bounding box and labeling the key-points at the major joints;

associating key-points of each participant across the images;

recovering a 3D body position of each of the plurality of participants based on the associated key-points; and

performing triangulation of a position of each of the major joints in a 3D space for a participant associated with the human joints, wherein the triangulation is performed based on camera projection matrices and pixel-locations of the major joints across the images.

9. The method of claim 8 , further comprising generating an animated model including animated motion for one or more of the plurality of participants.

10. The method of claim 9 , wherein associating the key-points of each participant across the images comprises:

determining a geometric pairwise correspondence between the images; and

determining one or more groups of images associated with a same participant from a plurality of camera angles.

11. The method of claim 10 , further comprising:

defining a region of interest within the images;

detecting each of the plurality of participants in the region of interest; and

generating a bounding box associated with each of the plurality of participants.

12. The method of claim 8 , further comprising:

performing kinematic model fitting to capture a position and orientation of one or more major segments of each of the plurality of participants to generate a model corresponding to each of the plurality of participants; and

performing temporal association to fit a model of each of the plurality of participants across a plurality of video frames.

13. At least one non-transitory computer readable medium having instructions stored thereon, which when executed by one or more processors, cause the processors to:

receive a plurality of two dimensional (2D) images captured by a camera array during a live event;

locate key-points of human joints of a plurality of event participants included in the images, wherein locating the key-points of the human joints comprises detecting major joints of each of the plurality of participants associated with a bounding box and labeling the key-points at the major joints;

associate key-points of each participant across the images;

recover a three-dimensional (3D) body position of each of the plurality of participants based on the associated key-points; and

performing triangulation of a position of each of the major joints in a 3D space for a participant associated with the human joints, wherein the triangulation is performed based on camera projection matrices and pixel-locations of the major joints across the images.

14. The non-transitory computer readable medium of claim 13 , wherein associating the key-points of each participant across the images comprises:

determining a geometric pairwise correspondence between the images; and

determining one or more groups of images associated with a same participant from a plurality of camera angles.

15. The non-transitory computer readable medium of claim 14 , having instructions stored thereon, which when executed by one or more processors, further cause the processors to:

define a region of interest within the images;

detect each of the plurality of participants in the region of interest; and

generate a bounding box associated with each of the plurality of participants.

16. The non-transitory computer readable medium of claim 13 , having instructions stored thereon, which when executed by one or more processors, further cause the processors to:

perform kinematic model fitting to capture a position and orientation of one or more major segments of each of the plurality of participants to generate a model corresponding to each of the plurality of participants; and

perform temporal association to fit a model of each of the plurality of participants across a plurality of video frames.