Abstract: [Objective] Improving unmanned sailboats' autonomous collision avoidance ability is especially important to ensure navigation safety. The local dynamic collision avoidance algorithm based on the improved fast marching square method is studied to solve sailboats' multi-vessel collision avoidance problem in open water. [Methods] Considering the uncontrollable speed of the sailboat, the collision avoidance behavior is transformed into a heading control problem, with the direction of the total potential field gradient decreasing as the desired collision avoidance heading. When sailing without risk of collision or downwind, construct a temporal potential field to avoid all stationary irregular obstacles and reach the target point. When the sailboat has sailing constraints, Gaussian likelihood functions are designed for specific encounter scenarios to dynamically construct obstacle potential fields according to the Convention on the International Regulations for Preventing Collisions at Sea (COLREGS), while the local wind potential field is introduced to consider the dead zone of a sailboat sailing to realize the application requirements of combining dynamic collision avoidance and zigzag sailing strategy for the sailboat. [Results] The simulation results show that the proposed algorithm can enable the unmanned sailboat to successfully achieve collision avoidance operations with other sailboats and restricted motor vessels in various encounter scenarios while complying with collision avoidance rules and avoiding navigational dead zones. Compared with the original fast-marching square algorithm, the upwind sailing time is significantly reduced, and its planned collision avoidance trajectory is more reasonable and safe. [Conclusion] The proposed method complies with the characteristics of sailboat motion and collision avoidance rules has high safety and robustness in complex environments, and is of scientific value to the development of autonomous obstacle avoidance technology for unmanned sailboats.