The invention provides a falling action detection method based on key skeleton point track analysis. The falling action detection method comprises the following steps of 1, firstly, collecting a plurality of groups of image sequences including the falling actions and a plurality of groups of image sequences including other actions except the falling actions; 2, extracting the key skeleton points, extracting the related information of the key skeleton points of the human body from the positive and negative samples, wherein the related information comprises the position information and the depth information of each skeleton point in the RGB image; 3, constructing a feature model, and generating a feature descriptor based on a key skeleton point track; and 4, constructing a classifier, classifying the track feature descriptors, and detecting a falling action. According to the present invention, the human skeleton position information is extracted by utilizing the OpenPose human skeleton detection algorithm based on the neural network, the most persuasive human skeleton points can be directly obtained, the method is not limited to the specific conditions and special environments, a user does not need to wear anything, the complex installation is not needed, and the cost is very low.