The invention discloses a motion classification and counting method, which is characterized by comprising the following steps of: firstly, carrying out data acquisition on various motions of differentpeople, and respectively marking the different motions; secondly, constructing a skeleton diagram for the time sequence frame skeleton information of the shot video stream, then performing convolution operation on the constructed skeleton diagram by utilizing a constructed time sequence model and a space model, and performing feature classification through comparison of data acquired in advance so as to identify different motion types; and then counting the exercise amount. The invention has the advantages that the method does not limit the motion type of the human body, does not have constraint and discomfort in the action process of the human body, is simple in structure, high in precision, low in cost, small in size, low in power consumption and convenient to use, has wide applicationin many fields, and has high practical value and application prospects.