〇HL Liu, T Taniguchi, Feature extraction and pattern recognition for human motion by a deep sparse autoencoder, The 14th IEEE International Conference on Computer and Information Technology (CIT 2014), pp. 173--181, (September 11-13, 2014 @Xi'an, Shaanxi, China)

发布日期：Jun 23, 2014 2:44:24 PM

Abstract:

Human motion data is high-dimensional time-series data, and it usually contains measurement error and noise. Recognizing human motion on the basis of such high-dimensional measurement row data is often difﬁcult and cannot be expected for high generalization performance. To increase generalization performance in a human motion pattern recognition task, we employ a deep sparse autoencoder to extract low-dimensional features, which can efﬁciently represent the characteristics of each motion, from the high-dimensional human motion data. After extracting low-dimensional features by using the deep sparse autoencoder, we employ random forests to classify low-dimensional features representing human motion. In experiments, we compared using the row data and three types of feature extraction methods - rincipal component analysis, a shallow sparse autoencoder, and a deep sparse autoencoder - for pattern recognition.

The experimental results show that the deep sparse autoencoder outperformed the other methods with the highest averaged accuracy rate, 75.1%, and the lowest standard deviation, ± 3.30%. The proposed method, application of a deep sparse autoencoder, thus enabled higher accuracy rate, better generalization and more stability than could be achieved with the other methods.

(Correction: In this paper, there is a slip of a pen, that is the dimensions of the input data was not 64, but 62. This mistake did not affect the results)

Page updated

Google Sites

Report abuse