Determination of a locational position for a camera to capture a collision of two or more actors
A computer-implemented method according to one embodiment includes running an initial network on a plurality of images to detect actors pictured therein and body joints of the detected actors. The method further includes running fully-connected networks in parallel, one fully-connected network for each of the detected actors, to reconstruct complete three-dimensional poses of the actors. Sequential model fitting is performed on the plurality of images. The sequential model fitting is based on results of running the initial network and the fully-connected networks. The method further includes determining, based on the sequential model fitting, a locational position for a camera in which the camera has a view of a possible point of collision of two or more of the actors. The camera is instructed to be positioned in the locational position.
1. A computer-implemented method, comprising:
running an initial network on a plurality of images to detect actors pictured therein and body joints of the detected actors;
running fully-connected networks in parallel, one fully-connected network for each of the detected actors, to reconstruct complete three-dimensional poses of the actors;
performing sequential model fitting on the plurality of images, wherein the sequential model fitting is based on results of running the initial network and the fully-connected networks;
determining, based on the sequential model fitting, a locational position for a camera in which the camera has a view of a possible point of collision of two or more of the actors; and
instructing the camera to be positioned in the locational position.
2. The computer-implemented method of claim 1 , wherein the determination of the locational position for the camera is based on a training model, wherein the training model is based on a second plurality of images captured by a camera positioned in a second locational position with a view of a collision of a plurality of other actors depicted in the second plurality of images.
3. The computer-implemented method of claim 1 , wherein the initial network is a convolutional neural network.
4. The computer-implemented method of claim 3 , wherein the convolutional neural network includes a plurality of two-dimensional kernels and a plurality of three-dimensional kernels, wherein at least some of the two-dimensional kernels are applied to determine the body joints of the detected actors, wherein at least some of the three-dimensional kernels are applied to determine a number of layers that exist in the plurality of images.
5. The computer-implemented method of claim 1 , wherein performing sequential model fitting on the plurality of images includes using kinematic skeleton fitting to obtain temporally coherent motion capture results.
6. The computer-implemented method of claim 5 , wherein the temporally coherent motion capture results are selected from the group consisting of: localizations of the actors relative to the camera, a joint angle parameterization of the actors, and a possible point of collision of two or more of the actors.
7. The computer-implemented method of claim 5 , wherein determining, based on the sequential model fitting, the locational position for the camera in which the camera has a view of the possible point of collision of two or more of the actors includes: applying the coherent motion capture results into a predetermined regression network.
8. A computer program product, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions readable and/or executable by a computer to cause the computer to:
run, by the computer, an initial network on a plurality of images to detect actors pictured therein and body joints of the detected actors;
run, by the computer, fully-connected networks in parallel, one fully-connected network for each of the detected actors, to reconstruct complete three-dimensional poses of the actors;
perform, by the computer, sequential model fitting on the plurality of images, wherein the sequential model fitting is based on results of running the initial network and the fully-connected networks;
determine, by the computer, based on the sequential model fitting, a locational position for a camera in which the camera has a view of a possible point of collision of two or more of the actors; and
instruct, by the computer, the camera to be positioned in the locational position.
9. The computer program product of claim 8 , wherein the determination of the locational position for the camera is based on a training model, wherein the training model is based on a second plurality of images captured by a camera positioned in a second locational position with a view of a collision of a plurality of other actors depicted in the second plurality of images.
10. The computer program product of claim 8 , wherein the initial network is a convolutional neural network.
11. The computer program product of claim 10 , wherein the convolutional neural network includes a plurality of two-dimensional kernels and a plurality of three-dimensional kernels, wherein at least some of the two-dimensional kernels are applied to determine the body joints of the detected actors, wherein at least some of the three-dimensional kernels are applied to determine a number of layers that exist in the plurality of images.
12. The computer program product of claim 8 , wherein performing sequential model fitting on the plurality of images includes using kinematic skeleton fitting to obtain temporally coherent motion capture results.
13. The computer program product of claim 12 , wherein the temporally coherent motion capture results are selected from the group consisting of: localizations of the actors relative to the camera, a joint angle parameterization of the actors, and a possible point of collision of two or more of the actors.
14. The computer program product of claim 12 , wherein determining, based on the sequential model fitting, the locational position for the camera in which the camera has a view of the possible point of collision of two or more of the actors includes: applying the coherent motion capture results into a predetermined regression network.
15. A system, comprising:
a processor; and
logic integrated with the processor, executable by the processor, or integrated with and executable by the processor, the logic being configured to:
run an initial network on a plurality of images to detect actors pictured therein and body joints of the detected actors;
run fully-connected networks in parallel, one fully-connected network for each of the detected actors, to reconstruct complete three-dimensional poses of the actors;
perform sequential model fitting on the plurality of images, wherein the sequential model fitting is based on results of running the initial network and the fully-connected networks;
determine, based on the sequential model fitting, a locational position for a camera in which the camera has a view of a possible point of collision of two or more of the actors; and
instructing the camera to be positioned in the locational position.
16. The system of claim 15 , wherein the determination of the locational position for the camera is based on a training model, wherein the training model is based on a second plurality of images captured by a camera positioned in a second locational position with a view of a collision of a plurality of other actors depicted in the second plurality of images.
17. The system of claim 15 , wherein the initial network is a convolutional neural network.
18. The system of claim 17 , wherein the convolutional neural network includes a plurality of two-dimensional kernels and a plurality of three-dimensional kernels, wherein at least some of the two-dimensional kernels are applied to determine the body joints of the detected actors, wherein at least some of the three-dimensional kernels are applied to determine a number of layers that exist in the plurality of images.
19. The system of claim 15 , wherein performing sequential model fitting on the plurality of images includes using kinematic skeleton fitting to obtain temporally coherent motion capture results.
20. The system of claim 19 , wherein the temporally coherent motion capture results are selected from the group consisting of: localizations of the actors relative to the camera, a joint angle parameterization of the actors, and a possible point of collision of two or more of the actors.