The invention discloses a multi-target tracking method by taking the depth as a leading clue under a dynamic scene, and relates to a multi-target tracking method. The multi-target tracking method comprises the following steps of: A, acquiring the current frame from a binocular camera, and correcting a binocular image; B, classifying colors of the image, and thus obtaining color blocks; C, performing three-dimensional reconstruction on the scene, and thus obtaining a dense depth picture; D, clustering depth information in an expanded area of a target region of the former frame to obtain a depth classification; E, classifying super pixels in the region and partitioning a foreground from a background according to the depth classification obtained in the step D; F, partitioning the foreground extracted in the step E according to an appearance model of the former frame, and extracting a final foreground; G, if a tracker is in a shielded state, stepping into the step I, and if the tracker is not shielded, returning to the step A, otherwise, recording a target appearance model of the former frame by the tracker, and stepping into the step H; H, extracting a depth block which shields an original target, and setting the tracker to be in the shielded state; and I, searching the original target around the current shielded block, if the original target is found, resetting the tracker to be in a normal state, and returning to the step A.