LIDAR data collected over Kahuku Training Area on the north side of Oahu Island, Hawaii (2005) were studied to determine the effectiveness and utility of identifying trails hidden under forest canopies. The Kahuku area is primarily comprised of lowland mesic grasslands and forests, with a rugged mountainous terrain. Using Quick Terrain Modeler, 3-D interactive analysis was done to identify trails. Results are verified by significant ground truth. Trails were found with overall accuracies up to 91%. Trails with widths of less than 2.5 m were detected with accuracies up to 85%.
2. Visually inspect surface model to find linear surface features.
3. Use knowledge of trail features to distinguish trail depressions from natural drainage features.
The two graphs below show a height profile across a road as compared to across a natural drainage depression. Roads will appear flat across the traveled way, whereas drainage features will have a 'V-shape'.
Trail Natural Drainage
4. Crop areas containing trails as "Target", and areas without trails as "Control". The Target area is approximately 41,643 m2 and the Control area is 41,974 m2.
5. Randomly select Target and Control point coordinates from Target/Control areas. The number of points (157 Target, 104 Control) were chosen based on total geographic area of Target/Control. Points within 5 m of each other were prohibited.
6. Use coordinates for "Ground Truth" verification. Sites within a 7 m radius of a trail were classified as 'trail'. The 7 m buffer was included to allow for cropping errors, rounding of UTM coordinates, and GPS positioning errors.
While these high percentages are very encouraging, it is important to
note that they are somewhat skewed due to the presence of a large
maintenance road running through the site. If these points are removed
from the analysis, the overall accuracy drops to 85%.
The next step in our research will be to automate this process.
Acknowledgments
National Geospatial-Intelligence Agency (NGA)
Methods