Evaluation of LIDAR for Automating Recognition of Roads and Trails Beneath Forest Canopy

This thesis discusses the utility of evaluating Light Detection and Ranging (LiDAR) to automate the recognition of roads and trails beneath forest canopy on Digital Elevation Models (DEMs) for use in military and forestry applications. Data were analyzed from three separate locations, including low elevation mixed conifer Indian Creek Watershed in Trinity County, CA; High elevation mixed conifer Cold Creek Trailhead area in South Lake Tahoe, CA; and lowland mesic forests in Kahuku training area, Oahu, HI. LiDAR data were evaluated to extract a DEM from ground points and to build a point cloud object layer between the estimated ground and an Above Ground Level (AGL) defined limit of 1.8 meters. By comparison of this point cloud data with the terrain model, small corridors above the forest floor extracted using linear feature detection were recognized as potential roads or trails. The object layer was of limited value, due partly to point collection density issues, and understory density in the different forest types. When evaluated using statistical classification techniques, results produced were inconsistent in segregating trails and roads from non-trail regions. It was determined that automated classification of these regions utilizing this method was ineffective and remains unacceptable without further research.