Date of Award
Master of Science (MS)
Hoover , Adam
Gowdy , John
This thesis presents a novel method of floor segmentation from a single image for mobile robot navigation. In contrast with previous approaches that rely upon homographies, our approach does not require multiple images (either stereo or optical flow). It also does not require the camera to be calibrated, even for lens distortion. The technique combines three visual cues for evaluating the likelihood of horizontal intensity edge line segments belonging to the wall-floor boundary. The combination of these cues yields a robust system that works even in the presence of severe specular reflections, which are common in indoor environments. The nearly real-time algorithm is tested on a large database of images collected in a wide variety of conditions, on which it achieves nearly 90% segmentation accuracy.
Additionally, we apply the floor segmentation method to low-resolution images and propose a minimalistic corridor representation consisting of the orientation line (center) and the wall-floor boundaries (lateral limit). Our study investigates the impact of image resolution upon the accuracy of extracting such a geometry, showing that detection of wall-floor boundaries can be estimated even in texture-poor environments with images as small as 16x12. One of the advantages of working at such resolutions is that the algorithm operates at hundreds of frames per second, or equivalently requires only a small percentage of the CPU.
Li, Yinxiao, "Segmentation of Floors in Corridor Images for Mobile Robot Navigation" (2011). All Theses. 1145.