A Collision Avoidance (CA) system with collision avoidance algorithms is critical to Autonomous Surface Vehicles (ASV). This paper explores a hybrid collision avoidance method based on an improved hybrid A* algorithm for inland waterway scenarios. This method uses discrete control actions, such as heading and speed setpoints, as input and generates a modified local path and control setpoints. A pre-processing method, namely a collision velocity check, is used to speed up the computation progress. Multiple aspects, including safety, path length, sailing time, rule compliance, and smoothness, are considered in finding the optimal collision-free path. We implemented the method under the Robot Operation System (ROS) architecture and conducted real-ship experiments for validation. The results show the effectiveness of this method.