Method and device for automatic gesture recognition

Patent metadata

Field	Value
Publication number	US-11475712-B2
Application number	US-201816615217-A
Country	US
Kind code	B2
Filing date	Apr 18, 2018
Priority date	Jun 1, 2017
Publication date	Oct 18, 2022
Grant date	Oct 18, 2022

Abstract

A method for automatic gesture recognition in which, by a machine learner data of a respective gesture of a variety of gestures executed by a user and captured by at least one sensor gestures of a class are assigned to a variety of predetermined classes, and in which the machine learner is trained with a training data set, which is divided into predetermined data segments. The respective data segments are data segments of the training data set are assigned to the variety of predetermined classes by means of the machine learner. A respective contribution, with the respective assignment processes for the assignment of respective data segments are incorporated into the automatic gesture recognition of a particular class.

First claim

The invention claimed is:

1. A method for the automatic recognition of gestures used to control a component of a vehicle comprising:
by a machine learner, data of a particular gesture of a variety of gestures carried out by a user and by at least one gesture captured by a particular class are assigned to a variety of predetermined classes,
wherein the machine learner is trained with a training data set, which is divided into predetermined data segments, wherein during the training respective data segments of the training data set are used by the machine learner assigned to the multitude of predetermined classes, wherein a respective contribution, with which respective assignment processes for assigning the respective data segments to a respective class are incorporated into the automatic gesture recognition, is taken into account by at least one weighting factor, wherein the at least one weighting factor to the weighting of the contribution, with the respective assignment processes for assigning the respective data segments to a respective class in the automatic gesture recognition is taken into account as a reciprocity in a loss function that optimizes the machine learner during training, with at least one weighting factor dynamically during the training of the machine learner is updated, and
where the loss function has respective assignment operations to map the respective data segments to a particular class using the at least one, depending on a frequency of assignment operations, from the respective data segments to a weighting factor in each class, and
where a control command associated with the respective class is generated and used to control the component of the vehicle, and where the weighting factor is one to the frequency of assignment operations of the respective data segments to the respective class proportional factor.

2. The method according to claim 1 , wherein respective weighting factors respective data segments to respective classes are normalized according to a class to which the least data segments are assigned.

3. The method according to claim 1 , wherein the machine learner is optimized by a local minimum of the loss function.

4. The method according to claim 1 , wherein an artificial neural network is chosen as a machine learner.

5. The method according to claim 1 , wherein the machine learner after a training automatically divides the data collected by at least one sensor into data segments and the data segments respective predetermined class.

6. A gesture sensing system for a vehicle, comprising:
at least one sensor for capturing gestures provided by a user and a control unit, wherein the control unit is configured to use a machine learner data of a particular gesture of a variety of gestures of a given class of predetermined gestures carried out by a user and captured by at least one sensor, and
class, and where the control unit is still configured to train the machine learner with a training data set divided into predetermined data segments, and during the training respective data segments of the training data set by the machine learner to assign to each class of the variety of predetermined classes, and whereby the control unit is still configured to use at least one weighting factor to weight a contribution, with the respective assignment processes for assigning the respective data segments to the respective classes are included in the automatic gesture recognition, and at least one weighting factor is taken into account as a reversing value in a loss function, with which the machine learner is optimized during training, with at least one weighting factor being dynamically updated during machine learner training, and the control unit is configured to have a weighting factor assigned to the respective class to generate a control command and use it to control a component of the vehicle, and where at least one weighting factor is a factor proportional to the frequency of assignment operations of each data segment to the respective class.

Assignees

• Audi Ag

Inventors

• Schön Torsten
• Schlittenbauer Michael
• Morgott Michael

Classifications

• G06V40/20 Primary

  Movements or behaviour, e.g. gesture recognition (recognition of facial expressions G06V40/16) · 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

• G06F18/2431

  Multiple classes · CPC title

• G06F18/214

  Generating training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title

• G06N3/084

  Backpropagation, e.g. using gradient descent · CPC title