As the problem of population aging is becoming more and more serious, a method of human fall detection based on wearable device is proposed to solve the social problem that the elderly are prone to fall. Different from the majority of fall detection methods which detect fall events after falling to the ground, the features of acceleration and angle are considered and support vector machine (SVM) is used as the classification algorithm to detect fall events before falling to the ground. The experiment results show that the fall event is recognized with a 99.2% recognition rate and the recognition rate of the activity of daily living is 96%. The average lead-time is 273ms.