Estimating egomotion from a video sequence is intrinsically difficult and requires high-level mathematics and programming skills. This work exploits existing technology to leverage the development of a mobile robot's navigation. An open source software MPEG encoder package is modified so that its motion vectors and encoded frame type are accessible. As a result, the process of estimating a motion field from the MPEG motion vectors is far less complicated and time-consuming than those used in conventional methods. The main contribution is the creation of low-cost multiple sensors for a mobile robot. Two real-time applications, visual odometry and precipice detection, are presented. Despite employing simple trigonometry, the visual odometry performs consistently well on moderately textured surfaces with low specular reflection. A proposed novel approach to detecting a precipice in real-time is shown to be successful even when the robot runs at a very high speed. The experimental results substantiate the use of both applications in real situations. |