Novel human gesture classification technique using bispectrum-based features extracted from variations of electromagnetic field under influence of gestures
Tutkimustuotos › › vertaisarvioitu
|Julkaisu||Telecommunications and Radio Engineering|
|DOI - pysyväislinkit|
|Tila||Julkaistu - 2017|
In this paper, novel human gesture detection, recognition and classification technique using gesture interaction with high-frequency electromagnetic field is suggested and experimentally studied. Suggested strategy is based on extraction of novel class of bispectrum-based information features from high-frequency electromagnetic field perturbed by human gestures. It is shown that phase bispectrum (biphase) contains information about the shape of radio frequency signal envelope resulted by signal disturbance with human gesture. Two novel kinds of classification features are proposed: first, sequence of limited biphase samples contained only in the radial slice in bispectral plane and, second, biphase signature contained array of biphase samples located within the main triangular bispectral domain. Performance of the bispectrum-based information features is studied experimentally for solving human gesture classification problem. Experimental results obtained by developed hardware and designed software are represented and discussed. It has been shown that proposed technique is invariant to random signal delays and variations of signal magnitude commonly observed in indoor multi-path electromagnetic field interference environment. It has been demonstrated by confusion matrix computations that human gesture classification performance evaluated for biphase values contained in the radial slice is worsen as compared to biphase signatures.