Search

Action recognition is usually a central problem for many practical applications, such as video annotations, video surveillance and human computer interaction. Most action recognition approaches are based on localized spatio-temporal features that can vary significantly when the viewpoint changes. However, their performance rapidly drops when the viewpoints of the training and testing data are different. In this paper, we propose a transfer learning framework for view-invariant action recognition by the way of sharing image stitching feature among different views. Experimental results on multi-view action recognition IXMAS dataset demonstrate that our method produces remarkably good results and outperforms baseline methods.

By study and simulation on the theoretic model and principle of electronic nose system in artificial olfactory system, this paper proposes an electronic nose system which combines gas sensor array and artificial neural network recognition algorithm. In order to testify systematic rationality and experimental realizability which is implemented in multiple gas measurement, we discuss related methods and steps which apply the system to quality and quantity of multiple gas. Through the analysis on experiments of gas detection which combines gas sensor array and pattern recognition technology, we apply BP feed forward neural network to analyze three gases in quality and quantity. Our research involves experimental configuration method of low concentration gas, input, output and hidden layer element determination method of BP feed forward network in quality and quantity analysis of multiple gases, signal pretreatment algorithm and gas sensor array response vector normalization. The experimental data show that the system designed in this paper makes recognition of 3 gases with 100% qualitative recognition and lower error of quantitative recognition.

Voiceprint Identification is a technology which can determine the identity of the speaker based on the voice. Now a good recognition performance can be obtained in the lab environment, but in the real environment, it is susceptible to the interference of ambient noise and a variety of channels, and the recognition rate declines. In order to improve the recognition rate, it's introduced the traditional voice denoising algorithms in the signal space, which can be applied to improve the signal to noise ratio(SNR). But when the noise is strong, the effect becomes minor, so the denoising method based on independent component analysis(ICA) is discussed in this paper, and experiments show that this denoising method improves SNR significantly.

Structurized video representation and indexing based on the representation are the foundation of video retrieval based on content. A novel algorithm is presented of automatically parsing and labelling video based on camera motion qualitative analysis. Firstly, the motion patterns are analyzed of the spatio-temporal slices in a video; and then the change detection and qualitative recognition of camera motion are carried out by using motion structure tensor. Finally, the results of camera motion analysis are used to parse and label video according to the camera motion coherence. The algorithm is experimentally proved effective and produces satisfactory results.