The　main　research　subject　in　this　paper　is　swimmer　detection　for　visual　surveillance　of　pool.　A　drowning　detection　method　based　on　background　subtraction　is　presented　in　this　paper.　The　consecutive　sequence　of　visual　surveillance　was　obtained　by　the　fixed　camera　installed　in　the　pool　wall.　Each　pixel　is　described　by　Gaussian　Mixed　Model,　set　up　self-adapted　background　model　and　updating　timely.　When　the　foreground　objects　was　separated,　for　getting　good　results,,　the　shadows　and　noises　must　be　removed.　The　experiments　results　show　that　this　method　is　effective　to　detect　the　drowners　and　eliminate　the　shadows.