In　the　background　of　global　aging,　more　attention　should　be　paid　to　the　elders＇　health　and　the　equality　of　their　life.　Nowadays　falls　became　one　of　greatest　danger　for　old　people.　Almost　62%　of　injury-related　hospitalizations　for　the　old　are　the　result　of　it.　In　this　paper,　we　propose　a　new　method　to　detect　fall　based　on　judging　human＇s　moving　posture　from　the　video.　It　consists　of　three　main　parts,　detecting　the　moving　object,　extracting　the　feature　and　recognizing　the　pattern　of　behavior.　To　improve　the　precision　and　increase　the　speed　of　the　detection,　we　adopt　two　layers　codebook　background　modeling　and　codebook　fragmentation　training.　Two　level　SVM　method　to　recognize　the　behavior:　In　the　first　level　of　the　SVM　classifier,　we　distinguish　the　standing　posture　and　other　posture　by　the　feature　of　moving　object,　such　as　the　ratio　of　the　major　and　minor　axis　of　the　ellipse.　In　the　second　level　of　the　SVM　classifier,　angle　of　the　ellipse　and　head　moving　trajectory　to　judge　the　falls　and　squat.　The　experimental　results　indicate　that　our　system　can　detect　fall　effectively.