Pedestrian Network Extraction from Fused Aerial Imagery (Orthoimages) and Laser Imagery (Lidar)

Abstract

A pedestrian network is a topological map that contains the geometric relationship between pedestrian path segments (e.g., sidewalk, crosswalk, footpath), which is needed in a variety of applications, such as pedestrian navigation services. However, current pedestrian networks are not widely available. In an effort to provide an automatic means for creating pedestrian networks, this paper presents a methodology for extracting pedestrian network from aerial and laser images. The methodology consists of data preparation and four steps: object filtering, pedestrian path region extraction, pedestrian network construction, and raster to vector conversion. An experiment, using ten images, was conducted to evaluate the performance of the methodology. Evaluation results indicate that the methodology can extract sidewalk, crosswalk, footpath, and building entrances; it collects pedestrian networks with 61 percent geometrical completeness, 67.35 percent geometrical correctness, 71 percent topological completeness and 51.38 percent topological correctness.