A special VO algorithm designed for the inland river environment is proposed to improve the performance. Firstly, the statistical properties of the feature of a typical maritime scene are collected and analyzed for the survival time and reprojection error at different areas in the image. Then, a water-shoreline detection (WSD) algorithm is adopted to divide the scene into two parts, i.e., water area and non-water area, discarding unstable features of the water surface. Finally, the Visual Odometry (VO) method is customized to generate navigation trajectories. The results demonstrate that this method is robust, and WSD improves the accuracy of trajectory estimation.