Automating identification of roads and trails under canopy using LiDAR

• Main Author: Harmon, Charles F.
• Other Authors: Olsen, Richard C.
• Published: Monterey, California. Naval Postgraduate School 2012
• Online Access: http://hdl.handle.net/10945/5585

Approved for public release; distribution is unlimited === Analysis techniques are developed to automatically extract roads and trails under thick forest canopy. LiDAR data were taken over the Swanton Pacific Ranch in the Santa Cruz Mountains from an airborne laser mapping system, the Optech 3100, on March 9-10, 2010. Collected data were characterized by point densities of 5-10 m2. Point cloud data were reduced to digital surface models using ARCMAP (from ESRI). The DSM was calculated at 1 meter spacing. These surface models were analyzed using topographic tools in ENVI, allowing for calculation of curvature, slope, convexity, and shaded relief. A multi-layer dataset was built and analyzed using spectral analysis tools in ENVI. The classification technique used was a combination of maximum likelihood classifier and a decision tree after use of erosion/dilation operators. Results are compared to ground truth collected in 2011. Classification resulted in 83.6% true positive rate, and the image processing result reduced the false positive rate to 3.0%.