We propose the use of ground surface segmentation to enhance the perception of obstacles in low to medium resolution prosthetic visual representations. We apply a recently proposed algorithm for segmenting traversable space in stereo disparity data, and show how such a scheme may be utilised to enhance the distinction between the ground surface and obstructions (in particular, small trip hazards). Qualitative comparisons with intensity and straight depth-based representations highlight advantages for the visualisation of obstacles, offering potential gains for visual navigation with low resolution and low dynamic range visual prostheses.