To get salient and reliable features is of great importance to robot navigation and other computer vision applications. This paper concentrates on feature detection, saliency description and matching for visual navigation. A corner detector based on chord-to-point distance accumulation is introduced to extract corners which represent the main structure of objects. Saliency descriptor of corner is defined according to its scale, angle, gradient, and rarity. Control points are selected according to the corners' saliency and tracked in sequential images with the method based on Fourier-Melline transform. Experiments show that the efficiency and robustness of vision navigation system are improved with the proposed method.