Institute for Information Industry

Chen-Kuo Chiang

Chih-Cheng Lin

Bo-Nian Chen

Yun-Zhong Lu

Yung-Pin Liu

Hao-Ting Li

Chih-Hsiang Yu

A classification model building apparatus and a classification model building method thereof are provided. The classification model building apparatus introduces a clustering algorithm to assist in training a deep learning model for classification and takes a sum of a clustering loss function, a center concentration loss function and a classification loss function multiplied by different weights, respectively, as a total loss function for training the deep learning model. Based on the total loss function, the classification model building apparatus adjusts parameters of the deep learning model through a backpropagation algorithm to build a classification model.

Classification model building apparatus and classification model building method thereof

An action evaluation model building apparatus and an action evaluation model building method thereof are provided. The action evaluation model building apparatus stores a plurality of raw data sets and a plurality of standard action labels corresponding thereto. Based on machine learning algorithms, the action evaluation model building apparatus computes the raw data sets and performs a supervised learning to build a feature vector creation model and a classifier model. The action evaluation model building apparatus determines a representation action feature vector of each standard action label by randomly generating a plurality of action feature vectors and inputting them into the classifier model. The action evaluation model building apparatus builds an action evaluation model based on the feature vector creation model, the classifier model and the representation action feature vectors.

Action evaluation model building apparatus and action evaluation model building method thereof

A motion recognition apparatus extracts a plurality of first motion feature vectors within a period of time, and extracts a plurality of second motion feature vectors within another period of time closely next to the period of time. The apparatus calculates and sequences the distances between each of the plurality of first motion feature vectors and the plurality of cluster centroids to generate a corresponding first cluster centroid sequence respectively, and calculates and sequences the distances between each of the plurality of second motion feature vectors and the plurality of cluster centroids to generate a corresponding second cluster centroid sequence respectively. The apparatus recognizes a first segment motion, a second segment motion, and a whole motion based on the number of appearances of the cluster centroid in the first cluster centroid sequences, the second cluster centroid sequences and both of them respectively.

Motion recognition apparatus and method

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.

Growing community of inventors

Taipei, Taiwan

Chen-Kuo Chiang