Body posture tracking

Patent metadata

Field	Value
Publication number	US-10830584-B2
Application number	US-201414288286-A
Country	US
Kind code	B2
Filing date	May 27, 2014
Priority date	May 29, 2013
Publication date	Nov 10, 2020
Grant date	Nov 10, 2020

Abstract

It is provided a computer-implemented method for performing body posture tracking, comprising the steps of collecting (S10) depth measurements of a body with at least one depth sensor; collecting (S20) inertial measurements with at least one inertial sensor attached to the body; and determining (S30) at least one posture of the body as a function of the depth measurements and the inertial measurements. Such a method improves the field of body posture tracking.

First claim

The invention claimed is:

1. A method for performing body posture tracking, comprising:
in an offline stage of machine-learning of a first probability distribution and of a second probability distribution:
(a) providing, (i) at least one first inertial sensor attached to a first body taking postures, (ii) at least one first depth sensor detached from the first body taking postures, and (iii) a first computer system having a processor;
(b) by the first body, performing a motion that forms a first gesture, the first gesture comprising a first series of postures;
(c) in real-time and continuously during the first gesture of the first body:
tracking each respective posture of the first series of postures,
by the at least one first inertial sensor, collecting inertial measurements of the first body,
by the at least one first depth sensor, collecting depth measurements of the first body, and
by the processor of the first computer system, learning the first probability distribution and the second probability distribution, the learning comprising:
constructing a training dataset comprising depth measurements of the first body and inertial measurements of the first body; and
by using the training dataset:
 updating parameters of the first probability distribution, the updating of the parameters comprising, for each respective posture of the first series of postures, comparing the tracked respective posture with one or more inertial measurements of the first body, thereby training the first probability distribution to assign probabilities to postures of a moving body as a function of inertial measurements of the moving body;
 updating parameters of the second probability distribution, the updating of the parameters comprising, for each respective posture of the first series of postures, comparing the tracked respective posture with one or more depth measurements of the first body, thereby training the first probability distribution to assign probabilities to postures of a moving body as a function of depth measurements of the moving body;

in an online stage of body posture tracking:
(a) providing (i) at least one second inertial sensor attached to a second body, (ii) at least one second depth sensor detached from the second body, (iii) a second computer system having a processor, and (iv) the first probability distribution and the second probability distribution;
(b) by the second body, performing a motion that forms a gesture, the gesture comprising a second series of postures, the second series of postures comprising one or more first postures and one or more second postures, the one or more first postures occurring before the one or more second postures occur;
(c) in real-time and continuously during the gesture:
by the at least one second depth sensor, collecting depth measurements of the second body;
by the at least one second inertial sensor, collecting inertial measurements of the second body; and
by the processor of the second computer system, determining the series of postures, the determining of the second series of postures comprising, for each respective posture of the second series of postures:
determining the respective posture, the determining of the respective posture comprising determining and maximizing a probability distribution that assigns probabilities to postures of the second body as a function of depth and inertial measurements of the second body collected at a current time of the respective posture, the probability distribution being recurrently obtained by multiplying the probability distribution that assigns probabilities to postures of the second body as a function of depth and inertial measurements collected at a previous time by the first probability distribution taking as input inertial measurements of the second body collected at the current time and by the second probability distribution taking as input depth measurements of the second body collected at the current time;

wherein the determining of the one or more first postures is performed before the one or more second postures occur.

2. The method of claim 1 , wherein determining a respective posture is repeated at each time that the method collects depth measurements and/or inertial measurements; and/or
determining a respective posture is repeated at each time that is a multiple of at least one predetermined time step;
the method thereby tracking body motion.

3. The method of claim 1 , wherein the depth measurements and the inertial measurements constitute time-series;
wherein determining a respective posture is repeated at each time that the method collects depth measurements and/or inertial measurements; and
wherein learning of the first probability distribution and of the second probability distribution comprises determining data based on which determining a respective posture at each current time is performed, the data including one or more of the depth measurements, the inertial measurements, and the current time of the respective posture.

4. The method of claim 3 , wherein the determined data comprise data based on which the method learns the first probability distribution and the second probability distribution.

5. The method of claim 1 , wherein determining the respective posture further comprises determining data based on which, if for the online stage, the depth measurements and the inertial measurements constitute time-series and determining whether the respective posture is repeated at each time the method collects depth measurements and/or inertial measurements, determining a posture at each current time may be performed.

6. The method of claim 5 , wherein the determined data comprise data based on which the method may learn the first probability distribution and/or the second probability distribution at the offline stage.

7. The method of claim 1 , wherein during the gesture, at least one posture of the second series of postures is such that a part of the body is occluded by another part of the body, each posture determined by the method being free of occlusion.

8. A non-transitory computer-readable medium configured to store instructions for performing body posture tracking, the instructions, when loaded and executed by a processor, causing the processor to:
in an offline stage of machine-learning of a first probability distribution and of a second probability distribution:
(a) providing, (i) at least one first inertial sensor attached to a first body taking postures, (ii) at least one first depth sensor detached from the first body taking postures, and (iii) a first computer system having a processor;
(b) by the first body, performing a motion that forms a first gesture, the first gesture comprising a first series of postures;
(c) in real-time and continuously during the first gesture of the first body:
tracking each respective posture of the first series of postures,
by the at least one first inertial sensor, collecting inertial measurements of the first body,
by the at least one first depth sensor, collecting depth measurements of the first body, and
by the processor of the first computer system, learning the first probability distribution and the second probability distribution, the learning comprising:
constructing a training dataset comprising depth measurements of the first body and inertial measurements of the first body; and
by using the training dataset:
 updating parameters of the first probability distribution, the updating of the parameters comprising, for each respective posture of the first series of postures, comparing the tracked respective posture with one or more inertial measurements of the first body, thereby training the first probability distribution to assign probabilities to postures of a moving body

Assignees

• Dassault Systemes

Inventors

• Boulkenafed Malika
• Michel Fabrice Francis

Classifications

• G06V10/811

  the classifiers operating on different input data, e.g. multi-modal recognition · CPC title

• G06F3/017 Primary

  Gesture based interaction, e.g. based on a set of recognized hand gestures (interaction based on gestures traced on a digitiser G06F3/04883) · CPC title

• G01B21/16 Primary

  for measuring distance of clearance between spaced objects · CPC title

• G06F18/256

  of results relating to different input data, e.g. multimodal recognition · CPC title

• G06V40/23

  Recognition of whole body movements, e.g. for sport training · CPC title