Estimation apparatus, learning apparatus, estimation method, learning method, and program

1. An estimation apparatus comprising:

an input data generation section that generates input data that contains region data indicating a position, a posture, or a motion about a region of a body and feature data indicating a feature of a time-series transition of

the position, the posture, or the motion about

the region data;

a feature extraction section that outputs new feature data in response to an input of the input data;

an estimation section that estimates a position, a posture, or a motion of another region of the body closer to a center of the body than the region on a basis of the feature data,

wherein the region data is made to correspond to a measurement result of the region,

wherein feature extraction section outputs feature data made to correspond to the latest measurement result that is based on the feature data previously output from the feature extraction section at a previous timing, and

wherein the estimation section estimates the position, the posture, or the motion of the other region of the body closer to the center of the body than the region at a timing of the latest measurement on a basis of the feature data indicating the feature of the time-series transition of the region data made to correspond to the measurement result until the latest measurement,

until a latest timing in response to the input of the input data that contains the region data at the latest timing and the feature data previously output from the feature extraction section and indicating the feature of the time-series transition at a timing preceding the latest timing, and the estimation section estimates the position, the posture, or the motion of the other region of the body closer to the center of the body than the region at the latest timing on the basis of the feature data indicating the feature of the time-series transition until the latest timing; and

a body tracking execution section that estimates a position and a direction of still another region of the body by inverse kinematics on a basis of an estimation result, by the estimation section, of the position, the posture, or the motion of the other region.

2. The estimation apparatus according to claim 1 , wherein

the region data is data indicating positions, postures, or motions about a plurality of regions different from one another.

3. The estimation apparatus according to claim 2 , wherein

the region data indicates a direction of a head included in the body, an angular speed of a left hand included in the body, and an angular speed of a right hand included in the body, and

the estimation section estimates a direction of a chest or a lumbar included in the body, or an angular speed of the chest or the lumbar included in the body.

4. The estimation apparatus according to claim 1 , wherein

the region data indicates a direction of a hand included in the body, an angular speed of the hand, and a position or a speed of the hand, and

the estimation section estimates a direction or an angular speed of a wrist of the hand included in the body.

5. An estimation apparatus comprising:

an input data generation section that generates input data that contains region data indicating a position, a posture, or a motion about a region of a body and feature data indicating a feature of a time-series transition of the region data;

a feature extraction section that outputs new feature data in response to an input of the input data;

an estimation section that estimates a position, a posture, or a motion of another region from the region on a basis of the feature data, wherein

wherein the region data is made to correspond to a measurement result of the region,

wherein feature extraction section outputs feature data made to correspond to the latest measurement result that is based on the feature data previously output from the feature extraction section at a previous timing, and

wherein the estimation section estimates the position, the posture, or the motion of the other region of the body at a timing of the latest measurement on the basis of the feature data indicating the feature of the time-series transition of the region data made to correspond to the measurement result until the latest measurement;

wherein feature extraction section outputs feature data made to correspond to the latest measurement result that is based on the feature data previously output from the feature extraction section at a previous timing;

a body tracking execution section that estimates a position and a direction of still another region of the body by inverse kinematics on a basis of an estimation result, by the estimation section, of the position, the posture, or the motion of the other region.

6. A learning apparatus comprising:

an input data generation section that generates input data that contains region data indicating a position, a posture, or a motion about a region of a body and feature data indicating a feature of a time-series transition of the region data;

a feature extraction section that outputs new feature data in response to an input of the input data;

an estimation section that estimates a position, a posture, or a motion of another region of the body closer to a center of the body than the region on a basis of the feature data;

a learning section that executes learning of at least one of the feature extraction section or the estimation section on a basis of a result of the estimation,

wherein the region data is made to correspond to a measurement result of the region,

wherein feature extraction section outputs feature data made to correspond to the latest measurement result that is based on the feature data previously output from the feature extraction section at a previous timing,

wherein the learning section executes the learning on the basis of the result of the estimation, by the estimation section, of the position, the posture, or the motion of the other region closer to the center of the body than the region at a timing of latest measurement based on the feature data indicating the feature of the time-series transition of the region data made to correspond to the measurement result until the latest measurement,

wherein feature extraction section outputs feature data made to correspond to the latest measurement result that is based on the feature data previously output from the feature extraction section at a previous timing; and

a body tracking execution section that estimates a position and a direction of still another region of the body by inverse kinematics on a basis of an estimation result, by the estimation section, of the position, the posture, or the motion of the other region.

7. A learning apparatus comprising:

an input data generation section that generates input data that contains region data indicating a position, a posture, or a motion about a region of a body and feature data indicating a feature of a time-series transition of the region data;

a feature extraction section that outputs new feature data in response to an input of the input data; and

an estimation section that estimates a position, a posture, or a motion of another region of the body on a basis of the feature data; and

a learning section that executes learning of at least one of the feature extraction section or the estimation section on a basis of a result of estimation,

wherein the region data is made to correspond to a measurement result of the region,

wherein feature extraction section outputs feature data made to correspond to the latest measurement result that is based on the feature data previously output from the feature extraction section at a previous timing,

wherein the learning section executes the learning on the basis of the result of the estimation, by the estimation section, of the position, the posture, or the motion of the other region at a timing of latest measurement based on the feature data indicating the feature of the time-series transition of the region data made to correspond to the measurement result until the latest measurement, and

wherein feature extraction section outputs feature data made to correspond to the latest measurement result that is based on the feature data previously output from the feature extraction section at a previous timing; and

a body tracking execution section that estimates a position and a direction of still another region of the body by inverse kinematics on a basis of an estimation result, by the estimation section, of the position, the posture, or the motion of the other region.

8. An estimation method comprising:

generating input data that contains region data indicating a position, a posture, or a motion about a region of a body and feature data indicating a feature of a time-series transition of the region data;

outputting new feature data in response to an input of the input data; and

estimating a position, a posture, or a motion of another region of the body closer to a center of the body than the region on a basis of the feature data,

wherein the region data is made to correspond to a measurement result of the region;

wherein the outputting includes outputting feature data made to correspond to the latest measurement result that is based on the feature data previously output from the feature extraction section at a previous timing,

wherein the estimating includes estimating the position, the posture, or the motion of the other region of the body closer to the center of the body than the region at a timing of the latest measurement on the basis of the feature data indicating the feature of the time-series transition of the region data made to correspond to the measurement result until the latest measurement;

outputting feature data made to correspond to the latest measurement result that is based on the feature data previously output from the feature extraction section at a previous timing; and

a body tracking execution section that estimates a position and a direction of still another region of the body by inverse kinematics on a basis of an estimation result, by the estimation section, of the position, the posture, or the motion of the other region.

9. A learning method comprising:

generating input data that contains region data indicating a position, a posture, or a motion about a region of a body and feature data indicating a feature of a time-series transition of the region data;

outputting, by the feature extraction section, new feature data in response to an input of the input data;

estimating, by an estimation section, a position, a posture, or a motion of another region of the body closer to a center of the body than the region on a basis of the feature data; and

executing learning of at least one of the feature extraction section or the estimation section on a basis of a result of estimation, wherein

the region data is made to correspond to a measurement result of the region,

wherein feature extraction section outputs feature data made to correspond to the latest measurement result that is based on the feature data previously output from the feature extraction section at a previous timing, and

the learning includes executing the learning on the basis of the result of the estimation, by the estimation section, of the position, the posture, or the motion of the other region closer to the center of the body than the region at a timing of latest measurement based on the feature data indicating the feature of the time-series transition of the region data made to correspond to the measurement result until the latest measurement,

wherein feature extraction section outputs feature data made to correspond to the latest measurement result that is based on the feature data previously output from the feature extraction section at a previous timing; and

estimating a position and a direction of still another region of the body by inverse kinematics on a basis of an estimation result, by the estimation section, of the position, the posture, or the motion of the other region.

10. A non-transitory computer readable medium having stored thereon a program for a computer, comprising:

generating, by an input data generation section, input data that contains region data indicating a position, a posture, or a motion about a region of a body and feature data indicating a feature of a time-series transition of the region data;

outputting, by a feature extraction section, new feature data in response to an input of the input data; and

estimating, by an estimation section, a position, a posture, or a motion of another region of the body closer to a center of the body than the region on a basis of the feature data, wherein

the region data is made to correspond to a measurement result of the region,

wherein feature outputting includes outputting feature data made to correspond to the latest measurement result that is based on the feature data previously output from the feature extraction section at a previous timing,

wherein the estimating includes estimating the position, the posture, or the motion of the other region of the body closer to the center of the body than the region at a timing of the latest measurement on the basis of the feature data indicating the feature of the time-series transition of the region data made to correspond to the measurement result until the latest measurement, and

wherein the outputting further includes outputting feature data made to correspond to the latest measurement result that is based on the feature data previously output from the feature extraction section at a previous timing; and

estimating a position and a direction of still another region of the body by inverse kinematics on a basis of an estimation result, by the estimation section, of the position, the posture, or the motion of the other region.

11. A non-transitory computer readable medium having stored thereon a program for a computer, comprising:

generating, by an input data generation section, input data that contains region data indicating a position, a posture, or a motion about a region of a body and feature data indicating a feature of a time-series transition of the region data;

outputting, by the feature extraction section, new feature data in response to an input of the input data;

estimating, by an estimation section, a position, a posture, or a motion of another region of the body closer to a center of the body than the region on a basis of the feature data; and

executing, by a learning section, learning of at least one of the feature extraction section or the estimation section on a basis of a result of estimation,

wherein the region data is made to correspond to a measurement result of the region,

wherein feature extraction section outputs feature data made to correspond to the latest measurement result that is based on the feature data previously output from the feature extraction section at a previous timing, and

wherein the learning includes executing the learning on the basis of the result of the estimation, by the estimation section, of the position, the posture, or the motion of the other region closer to the center of the body than the region at a timing of latest measurement based on the feature data indicating the feature of the time-series transition of the region data made to correspond to the measurement result until the latest measurement,

wherein feature extraction section outputs feature data made to correspond to the latest measurement result that is based on the feature data previously output from the feature extraction section at a previous timing; and

estimating, by a body tracking execution section, a position and a direction of still another region of the body by inverse kinematics on a basis of an estimation result, by the estimation section, of the position, the posture, or the motion of the other region.