The invention relates to a continuous tracking method suitable for target grabbing of an underwater robot, and belongs to the technical field of visual target tracking. The method comprises the stepsof establishing a kernel-related motion model; performing HOG feature extraction; establishing an appearance model; and updating and iterating the model. According to the invention, based on the kernel correlation filtering theory, dense sampling is realized through cyclic displacement of base samples, so that a richer training sample set is extracted, HOG features are extracted, the appearance model of a tracking target is established, a kernel function is introduced, so that the nonlinear regression problem is solved, the calculation efficiency is improved, whether reinitialization trackingis needed or not is judged according to a feedback result, a self-discrimination mechanism based on system confidence is provided, and continuous tracking of the target is realized. According to the method, stable tracking of the underwater target can be guaranteed, shielding and mistaken tracking conditions can be automatically judged, so that re-identification and re-tracking are realized, and continuous tracking and reliable grabbing of the underwater target are further completed.