Human　activity　recognition　is　one　of　the　core　technologies　of　human-centered　Internet　of　Things.　In　order　to　create　a　cost-efficient　activity　recognition　system　without　carrying　any　wearable　devices,a　human　activity　recognition　method　based　on　WiFi　channel　state　information(CSI)is　proposed.　This　method　first　filters　the　collected　CSIs　with　Hampel　filtering　and　discrete　wavelet　denoising,then　obtains　the　starting　and　ending　time　of　the　activity　by　detecting　the　variance　of　CSI,thus　extracts　the　feature　vector　from　the　CSI　of　the　corresponding　time　period　and　identifies　human　activity　by　using　linear　discriminant　analysis(LDA)classifier.　The　experiments　show　that　the　average　recognition　rate　of　the　five　daily　movements:squatting,standing,sitting,walking　and　picking　up　can　reach　96%.　©　2019,　The　Editorial　Office　of　Chinese　Journal　of　Sensors　and　Actuators.　All　right　reserved.