Disclosed is a dynamic collision avoidance method for an unmanned surface vessel based on route replanning. The method comprises the following steps: acquiring navigation information and pose information of a neighboring ship of an unmanned vessel itself via a vessel-borne sensor; constructing a collision cone between the unmanned vessel and the neighboring ship; introducing a degree of uncertainty with respect to observing movement information of the neighboring ship and applying a layer of soft constraint to the collision cone; applying a speed and a heading limit range of the unmanned vessel; acquiring an ultimate candidate speed set; introducing a cost function to select an optimum collision avoidance speed; and performing an internal recycle of navigation simulation with the optimum collision avoidance speed to obtain a route replanning point for dynamic collision avoidance of the unmanned vessel. According to the present invention, a dynamic collision avoidance strategy of the unmanned surface vessel is output in form of route replanning to meet constraints of international regulations for preventing collisions at sea, and it is well adapted to manipulate and control the unmanned vessel itself, so that a dynamic collision avoidance requirement of the unmanned vessel is met.