Identification system and identification method

Patent metadata

Field	Value
Publication number	US-10185870-B2
Application number	US-201715499878-A
Country	US
Kind code	B2
Filing date	Apr 27, 2017
Priority date	Mar 3, 2017
Publication date	Jan 22, 2019
Grant date	Jan 22, 2019

Abstract

An identification method includes: sensing movement data; capturing multiple feature data from the movement data; cutting the first feature data into a plurality of first feature segments, dividing the first feature segments into a plurality of first feature groups, and calculating multiple first similarity parameters of the first feature groups respectively corresponding to a plurality of channels; making the first feature groups correspond to the channels according to the first similarity parameters; simplifying the first feature groups corresponding to the channels respectively by a convolution algorithm to obtain a plurality of first convolution results corresponding to the first feature groups; simplifying the first convolution results corresponding to the first feature groups respectively by a pooling algorithm to obtain multiple first pooling results corresponding to the first feature groups; and combining the first pooling results corresponding to the first feature groups to generate a first feature map.

First claim

What is claimed is:

1. An identification system, comprising:
a processor for receiving movement data from a sensing device, wherein the processor comprises:
a preprocessing module for capturing a plurality of feature data from the movement data, wherein the feature data comprise a first feature data and a second feature data;
a data cutting module for cutting the first feature data into a plurality of first feature segments, dividing the first feature segments into a plurality of first feature groups and calculating a plurality of first similarity parameters of the first feature groups respectively corresponding to a plurality of channels;
a channel matching module for making the first feature groups correspond to the channels according to the first similarity parameters;
a data integration module for simplifying the first feature groups corresponding to the channels respectively by a convolution algorithm to obtain a plurality of first convolution results corresponding to the first feature groups, simplifying the first convolution results corresponding to the first feature groups respectively by a pooling algorithm to obtain a plurality of first pooling results corresponding to the first feature groups, combining the first pooling results corresponding to the first feature groups to generate a first feature map, substituting the first feature map and a second feature map into the convolution algorithm again to obtain a second convolution result, and substituting the second convolution result into the pooling algorithm again to obtain a second pooling result; and
a model training module for substituting the second pooling result into a model training algorithm to generate an identification model.

2. The identification system of claim 1 , wherein the data cutting module further is used for cutting the first feature data into the first feature segments according to a time sequence and dividing the first feature segments into groups according to the time sequence to obtain the first feature groups.

3. The identification system of claim 2 , wherein the time sequence refers to the sequential order of generating the first feature segments in the first feature data.

4. The identification system of claim 1 , wherein the data cutting module cuts the second feature data into a plurality of second feature segments and divides the second feature segments into groups to obtain a plurality of second feature groups;
the channel matching module calculates a plurality of second similarity parameters of the second feature groups respectively corresponding to the channels, and makes the second feature groups correspond to the channels according to the second similarity parameters; and
the data integration module simplifies the second feature groups corresponding to the channels respectively by the convolution algorithm to obtain the first convolution results corresponding to the second feature groups, simplifies the first convolution results corresponding to the second feature groups respectively by the pooling algorithm to obtain the first pooling results corresponding to the second feature groups, and combines the first pooling results corresponding to the second feature groups to generate the second feature map.

5. The identification system of claim 1 , wherein the model training algorithm is at least one of a neural network algorithm, a support vector machine, a decision tree, a Bayesian classifier, an expectation maximization and a K-nearest neighbor, wherein when the model training algorithm is the neural network algorithm, the identification model is generated by means of at least one hidden layer and a classifier of the neural network algorithm.

6. The identification system of claim 1 , wherein after capturing the first feature data from the movement data, the preprocessing module determines a remaining data as the second feature data, and the remaining data is obtained by subtracting the first feature data from the movement data.

7. The identification system of claim 1 , wherein the movement data is an acceleration data, and the first feature data is a gravity data.

8. The identification system of claim 1 , wherein the data integration module is further used for generating a plurality of second pooling results, substituting the second pooling results into the convolution algorithm to generate a plurality of third convolution results, substituting the third convolution results into the pooling algorithm respectively to obtain a plurality of third pooling results.

9. The identification system of claim 8 , wherein the model training module generates the identification model according to the third pooling results.

10. The identification system of claim 1 , wherein the channel matching module calculates an average, a covariance, a distance value or a related coefficient value of a plurality of content values of each of the first feature groups, so as to obtain the first similarity parameters.

11. An identification method, comprising:
sensing movement data;
capturing a plurality of feature data from the movement data; wherein the feature data comprise a first feature data and a second feature data;
cutting the first feature data into a plurality of first feature segments, dividing the first feature segments into a plurality of first feature groups, and calculating a plurality of first similarity parameters of the first feature groups respectively corresponding to a plurality of channels;
making the first feature groups correspond to the channels according to the first similarity parameters;
simplifying the first feature groups corresponding to the channels respectively by a convolution algorithm to obtain a plurality of first convolution results corresponding to the first feature groups;
simplifying the first convolution results corresponding to the first feature groups respectively by a pooling algorithm to obtain a plurality of first pooling results corresponding to the first feature groups;
combining the first pooling results corresponding to the first feature groups to generate a first feature map;
substituting the first feature map and a second feature map into the convolution algorithm again to obtain a second convolution result;
substituting the second convolution result into the pooling algorithm again to obtain a second pooling result; and
substituting the second pooling result into a model training algorithm to generate an identification model.

12. The identification method of claim 11 , further comprising:
cutting the first feature data into the first feature segments according to a time sequence and dividing the first feature segments into the first feature groups according to the time sequence.

13. The identification method of claim 12 , wherein the time sequence refers to the sequential order of generating the first feature segments in the first feature data.

14. The identification method of claim 11 , further comprising:
cutting the second feature data into a plurality of second feature segments and dividing the second feature segments into a plurality of second feature groups;
calculating a plurality of second similarity parameters of the second feature groups respectively corresponding to the channels, and making the second feature groups correspond to the channels according to the second similarity parameters;
simplifying the second feature groups corresponding to the channels respectively by the convolution algorithm to obtain the first convolution results corresponding to the second feature groups;
simplifying the first convolution results corresponding to the second feature groups respectively by the pooling algor

Assignees

• Inst Information Ind

Inventors

• Chiang Chen-Kuo
• Yu Chih-Hsiang
• Chen Bo-Nian

Classifications

• G06V40/20

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

• G06N3/08 Primary

  Learning methods · CPC title

• G06F18/2414

  Smoothing the distance, e.g. radial basis function networks [RBFN] · CPC title

• G06F2218/12

  Classification; Matching · CPC title

• G06N3/045

  Combinations of networks · CPC title