This paper proposes to use optical flow to get visual feedback to control the navigation of a mobile robot using only the onboard computational setup. A very fast algorithm is proposed to compute the optical flow, and a confidence measure is applied to the resulting optical flow vectors to eliminate the more unreliable ones. An image segmentation process based on the optical flow is then implemented, which results in splitting the scene in different objects. Each object is then associated to a time to collision, whose values are used to control the angular and linear speeds of the mobile robot.