Homography-based ground area detection for indoor mobile robot using binocular cameras

Drivable area detection is an important issue for mobile robot to work in indoor and outdoor environments with safety and security with respect to itself and environment. In this paper, we propose an approach to detect the ground area for indoor mobile robots with binocular cameras. At first, the ground induced homography between onboard binocular cameras is calibrated. In ground area detection, the right eye image is warped to left eye by the homography and subtracted by the left eye image. From the difference image the rough non-ground regions can be detected. Then the closed contours on the difference image and the left eye image are searched to segment them into regions. These corresponding regions are compared to judge whether they belong to ground plane. The conducted experiments show that the proposed method is valid in detecting ground area.