In order to acquire fall information of old man quickly and effectively

propose a fast and effective recognition algorithm for human fall. The algorithm is based on depth sequence. It uses skeleton joint data provided by Kinect to construct model for human action representation. Human action is regarded as a set of joint motion curve and velocity curve. Discrete Fréchet distance is used as similarity metric between the joint motion curve and velocity curve. K nearest neighbor（ KNN） classifier is used to classify the action samples. Experiments on public SDUFall dataset show that the proposed method is better than the existing methods.