A monocular vision-based navigation algorithm for mobile robots moving in unknown environments was presented, which consists of obstacle detection, monocular vision based distance measurement and local path planning. In order to decrease the impact of lighting disturbance on obstacle detection, pixel-based image segmentation in HSI color space was used to get the set of the image coordinates of the obstacle boundaries by comparing histograms of current image window and reference image window. In order to realize distance measuring between obstacles and a robot, a transformation equation between the vision coordinates and the robot coordinates was deduced by geometrical reasoning. Grid map was used to model the rectangular area transformed from the trapezoidal field of view of the camera and grids with obstacles were marked according to the obstacle boundaries and the measured distance. With a new searching method a safe local path for the robot was found by searching the grid map. Experiments done in real environments show that less sensitive to highlight and shading, the proposed algorithm could correctly plan a safe local path to navigate the robot.