Indoor Mobile Robot Navigation by Center Following based on Monocular Vision

We address the problem of indoor mobile robot navigation by center following without prior environmental information based on visual information provided by a single camera. Recently, the research on the mobile robot of the automatic moving type works actively (DeSouza & Kak, 2002). It is an important problem to acquire moving environment information to move automatically. Various sensors such as the ultrasonic sensor, the position sensing device (PSD) sensor, the laser rangefinder, radar, and camera are used to acquire moving environmental information. The ultrasonic sensor is cheap but suffers from specular reflections and usually from poor angular resolution. The laser rangefinder and radar provide better resolution but is more complex and more expensive. These range-based sensors have difficulty detecting small or flat object on the ground. These sensors are also unable to distinguish between difference types of ground surfaces. While small objects and different types of ground are difficult to detect with range-based sensors, they can in many cases be easily detected with color vision. Though the obtained accuracy of distance using the camera decrease compared with the range-based sensors, various methods for acquiring the moving environment with one or more cameras are proposed. The stereo vision can measure distance information with two or more cameras as well as the ultrasonic sensor and the PSD sensor (Herath et al., 2006). However, the processing cost becomes complex with two or more cameras. The omni-directional camera has an advantage to obtain all surrounding environments of the robot at one time (Gaspar et al, 2000; Joochim & Chamnongthai, 2002; Argyros et al., 2002). However, the omni-directional camera is a special camera, and should mount it on the top of the robot to take all round view. This causes the limitation in appearance. Since the detection of obstacle region or the wall is difficult with the acquisition image, it is necessary to convert to the panoramic image. A lot of mobile robots with only one camera are proposed (Vassallo et al., 2000; Ebner & Zell, 2000; Tomono & Yuta, 2004; Aider et al., 2005; Hayashi, 2007; Bellotto et al., 2008). Using a single camera, only forward information can be acquired, and information is less than the stereo vision and omni-directional camera. However, if the robot moves to the forward, the means to supplement with other sensors, such as the ultrasonic sensor and PSD sensor is considered, even if the accurate intelligence is not acquired. Moreover, it has the advantage that the processing cost decreases compared with two or more cameras. O pe n A cc es s D at ab as e w w w .ite ch on lin e. co m