The invention discloses a human tracking method and a system, which are used for solving the problems that in a point cloud modeling, when the points of the blocked head, limbs and other parts do not exist, or the point cloud resolution is lower, how to build the position attitudes of the parts in a model. The method disclosed by the invention adopts a fitting algorithm to replace a matching algorithm to calculate the attitudes of the parts of the human body, and comprises the following steps of: firstly, calculating according to point cloud data and the attitude data of the human body model to obtain each point in a human body point cloud and paring of a part closest to the point, i.e., the closest pair of points; then calculating according to the closest pair of points to obtain four sidelines of a rectangle representing the trunk and a segment representing the head, and simultaneously adopting the fitting algorithm to calculate and obtained segments representing the limbs accordingto the closest part of points; and finally, according to the obtained four sidelines, and the segments respectively representing the head and the limbs, outputting the human body model. Accordingly, under extreme situations, the human body model can be better constructed, and the human motion can be analyzed and tracked by the method provided by the invention.