The invention discloses a tumble detection method for a human body based on a three-axis acceleration sensor, which acquires human body three-dimensional acceleration signals through the three-axis acceleration sensor. The method adopts the three-axis acceleration sensor to acquire the acceleration signals of the human body in three directions, adopts two characteristic quantities of SVM and SMA to detect the tumble of the human body. The method comprises the following particular steps: 1) acquiring acceleration data of the human body; 2) performing data preprocessing; 3) calculating SVM value; 4) performing SVM wavelet analysis; 5) detecting SVM detection, if the SVMs of a plurality of consecutively acquired sample points are larger than or equal to the threshold value th1, confirming the tumble of the human body initially, and entering step 6), otherwise, continuing step 5); 6) calculating SMA value, if the SMA is larger than or equal to the threshold value th2, entering step 7), otherwise, determining as suspicious tumble, returning to step 5); 7) if the SMAs of a plurality of consecutively acquired sample points is larger than or equal to the threshold value th2, confirming the occurrence of tumble finally and alarming, otherwise, determining as suspicious tumble, and returning to step 5).