With the aging of the deepening of the world, the fall accident of elder has been taken great attention by more and more people. This paper is committed to invent a set of automatic fall detection device, so as to reduce the damage for elder caused by fall accident and apply timely assistance. Therefore, we install a Tri-axial G-sensor on chest to acquire acceleration information, and establish a fall detection algorithm based on hidden Markov model. First the device can extract data features, then learn fall process to form a Markov fall model, finally, detect real-time data through the model to judge fall accidents from all the daily behavior. Experimental results show that the wearable device can effectively identify a simple fall process with high accuracy.