Detection of Human Activity using Cepstrum of Gas Sensor Responses

本報告は、ガスセンサにおける流体伝達場をモデル化し、室内環境の変化を解析する。1日の観測応答から複素ケプストラムのロータイムウインドによって伝達系を抽出し、推定空間伝達関数の逆フィルタリングにより人の行動に伴う活動量を復元する。これらを、一般家庭における実測実験にて確認し、本提案手法の有効性を示す。We modeled fluid transfer field at gas sensor responses, and analyed changes in the indoor environment. The results of this analysis indicate that transfer characteristics can be extracted by blind deconvolution using the low time windowing of complex cepstrum, and inverse filtration of the thus estimated space transfer characteristics allows us to extract drive signals proportional to the amounts of human activity. In an experiment in which the indoor environmental changes during daily life were measured, and drive signals were obtained. These results suggest that fhe detection of human actions can be improved by eliminating the residual components from the gas sensor responses