It is found that a timely detection of the physical falling of the elderly is an issue concerning the personal safety of the elderly. A judgement algorithm with SVM and SVM variation indices as the primary judgement, and angle data as the assistant judgement, has thus been proposed. In order to ensure the accuracy of data, the collected data are to be used for computing SVM after a processed window smoothing, thus obtaining the threshold value of SVM and SVM variation indices under falling state through experimental tests. The proposed method is characterized with such advantages as the omission of the integral operation in the process of judgment, and the filtration of the influence of other body postures on the physica falling detection by multilevel judgment and angle aided judgment.