Street model with multiple movable panels for pedestrian environment analysis

We propose a new pedestrian environment recognition method for pedestrian walk at a comparatively narrow street area. Pedestrian road side solid structures are modelled as pictures put on flat walls. In our approach, we use multiple, movable panels for modelling pedestrian walk space area. We adjust an opaque panel to road side solid structures and a map texture on the surface. For small structures such as guardrails, utility poles in the road area, we put these textures on a transparent panel. For estimating the wall position in the image, we use a vertical and horizontal line segment of the walls. We adjust the bottom line of the panel to a line connecting bottom points of vertical line segments of walls. Horizontal line segment of walls are used for constraining a search area. For transparent panel adjusting, we use systematic change in texture elements caused by projection. Our method deals with three types of road shapes, straight roads and intersection (+-shaped roads , T-shaped roads). Experimental results for still images show the effectiveness of the proposed method. It can be applied for a visually impaired system.