Continuous interaction learning and detection in real-time

Patent metadata

Field	Value
Publication number	US-10366345-B2
Application number	US-201615182295-A
Country	US
Kind code	B2
Filing date	Jun 14, 2016
Priority date	Mar 15, 2013
Publication date	Jul 30, 2019
Grant date	Jul 30, 2019

Abstract

Systems and methods may provide for partitioning a plurality of training samples into a first sequential list of centroids, removing one or more repeating centroids in the first sequential list of centroids to obtain a first reduced list of centroids and generating a set of Hidden Markov Model (HMM) parameters based on the first reduced list of centroids. Additionally, a plurality of detection samples may be partitioned into a second sequential list of centroids, wherein one or more repeating centroids in the second sequential list of centroids may be removed to obtain a second reduced list of centroids. The second reduced list of centroids may be used to determine a match probability for the plurality of detection samples against the set of HMM parameters. In one example, the reduced lists of centroids lack temporal variability.

First claim

I claim:

1. An apparatus to process training samples, comprising:
a computing device including a sensor to collect a plurality of training samples selected from human motion samples or human verbal samples, the computing device including:
a partition module, implemented at least partly in one or more of configurable logic or fixed-functionality logic hardware, to partition the plurality of training samples into a sequential list of centroids including non-repeating centroids and one or more repeating centroids, wherein the sequential list has a first of the non-repeating centroids, the one or more repeating centroids and a second of the non-repeating centroids in that stated order so that the first non-repeating centroid is to transition to the one or more repeating centroids, and the one or more repeating centroids are to transition to the second non-repeating centroid, wherein the one or more repeating centroids repeat the first non-repeating centroid such that the sequential list of centroids has a temporal dimension represented by the one or more repeating centroids;
a filter, implemented at least partly in one or more of configurable logic or fixed-functionality logic hardware, to remove the temporal dimension from the sequential list to obtain a reduced list of centroids lacking temporal variability, wherein to remove the temporal dimension to obtain the reduced list, the filter is to remove the one or more repeating centroids from the sequential list while each of the non-repeating centroids from the sequential list is retained so that each of the non-repeating centroids is in the reduced list and the first non-repeating centroid in the reduced list is to transition to the second non-repeating centroid in the reduced list;
a parameter module, implemented at least partly in one or more of configurable logic or fixed-functionality logic hardware, to generate a set of Hidden Markov Model (HMM) parameters based on the reduced list of centroids lacking the temporal variability; and
a detection module, implemented at least partly in one or more of configurable logic or fixed-functionality logic hardware, to determine a match probability for detection samples against the set of HMM parameters.

2. The apparatus of claim 1 , wherein the sequential list of centroids is to be associated with a set of clusters having a one-to-one mapping to states in the HMM.

3. The apparatus of claim 1 , wherein the partition module includes:
a cluster unit, implemented at least partly in one or more of configurable logic or fixed-functionality logic hardware, to determine a set of clusters for the plurality of training samples;
a centroid unit, implemented at least partly in one or more of configurable logic or fixed-functionality logic hardware, to determine a set of cluster centroids corresponding to the set of clusters; and
an assignment unit, implemented at least partly in one or more of configurable logic or fixed-functionality logic hardware, to assign each training sample in the plurality of training samples to a centroid in the set of cluster centroids to partition the plurality of training samples into the sequential list of centroids.

4. The apparatus of claim 1 , wherein centroids, that are associated with the detection samples, are subjected to a forward algorithm to determine the match probability against the set of HMM parameters.

5. A method to process training samples, comprising:
partitioning, with a computing device, a plurality of training samples selected from human movement samples or human verbal samples into a sequential list of centroids including non-repeating centroids and one or more repeating centroids, wherein the sequential list has a first of the non-repeating centroids, the one or more repeating centroids and a second of the non-repeating centroids in that stated order so that the first non-repeating centroid is to transition to the one or more repeating centroids, and the one or more repeating centroids are to transition to the second non-repeating centroid, wherein the one or more repeating centroids repeat the first non-repeating centroid such that the sequential list of centroids has a temporal dimension represented by the one or more repeating centroids;
modifying, with the computing device, the sequential list to remove the temporal dimension from the sequential list to obtain a reduced list of centroids lacking temporal variability, wherein the modifying to remove the temporal dimension includes removing the one or more repeating centroids from the sequential list while retaining each of the non-repeating centroids from the sequential list so that each of the non-repeating centroids is in the reduced list and the first non-repeating centroid in the reduced list is to transition to the second non-repeating centroid in the reduced list; and
generating, with the computing device, a set of Hidden Markov Model (HMM) parameters based on the reduced list of centroids lacking the temporal variability; and
determining a match probability for detection samples against the set of HMM parameters.

6. The method of claim 5 , wherein the sequential list of centroids is associated with a set of clusters having a one-to-one mapping to states in the HMM.

7. The method of claim 5 , wherein partitioning the plurality of training samples includes:
determining a set of clusters for the plurality of training samples;
determining a set of cluster centroids corresponding to the set of clusters; and
assigning each training sample in the plurality of training samples to a centroid in the set of cluster centroids to partition the plurality of training samples into the sequential list of centroids.

8. The method of claim 5 , wherein centroids, that are associated with the detection samples, are subjected to a forward algorithm to determine the match probability against the set of HMM parameters.

9. A method to process detection samples, comprising:
partitioning, with a computing device, a plurality of detection samples selected from human movement samples or human verbal samples into a sequential list of centroids including non-repeating centroids and one or more repeating centroids, wherein the sequential list has a first of the non-repeating centroids, the one or more repeating centroids and a second of the non-repeating centroids in that stated order so that the first non-repeating centroid is to transition to the one or more repeating centroids, and the one or more repeating centroids are to transition to the second non-repeating centroid, wherein the one or more repeating centroids repeat the first non-repeating centroid such that the sequential list of centroids has a temporal dimension represented by the one or more repeating centroids;
modifying, with the computing device, the sequential list to remove the temporal dimension from the sequential list to obtain a reduced list of centroids lacking temporal variability, wherein the modifying to remove the temporal dimension includes removing the one or more repeating centroids from the sequential list while retaining each of the non-repeating centroids from the sequential list so that each of the non-repeating centroids is in the reduced list and the first non-repeating centroid in the reduced list is to transition to the second non-repeating centroid in the reduced list; and
determining, with the computing device, a match probability for the plurality of detection samples against a set of Hidden Markov Model (HMM) parameters, that are associated with a training session, based on the reduced list of centroids lacking the temporal variability.

10. The method of claim 9 , wherein the sequential list of centroids is associated with a set of clusters having a one-to-one

Assignees

• Intel Corp

Inventors

• Evans Chuck

Classifications

• G06N7/01

  Probabilistic graphical models, e.g. probabilistic networks · CPC title

• G06F18/295

  Markov models or related models, e.g. semi-Markov models; Markov random fields; Networks embedding Markov models · CPC title

• G06F18/24137 Primary

  Distances to cluster centroïds · CPC title

• G06F18/214

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

• G06N20/00 Primary

  Machine learning · CPC title