Objective:　To　propose　a　gesture　recognition　algorithm　based　on　three-axis　acceleration　sensor　signal.　Methods:　To　extract　the　tri-axial　acceleration　signals　in　the　time　domain,　frequency　domain,　time-frequency　frequency　characteristics　of　the　design　of　the　SVM　classifiers　on　a　publicly　available　database　of　5　class　actions　(relaxing,　jumping,　walking,　marking　time,　running)　classification.　Results:　The　average　recognition　rate　of　the　5　motions　is　94.37%,　and　the　jumping　and　static　identification　rate　are　98.13%　and　96.60%　respectively.　Conclusion:　The　feature　extraction　of　the　acceleration　signal　of　8s　can　be　applied　to　real-time　system　analysis.