hf

Uploaded from authorPOINTLite
Views:
 
Category: Entertainment
     
 

Presentation Description

No description available.

Comments

Presentation Transcript

Automatic Tree-Crown Segmentation Using LM Filters and Balloons: 

Automatic Tree-Crown Segmentation Using LM Filters and Balloons Bert Rawert Advisors: Howard and Ed

Why? : 

Why? Automatic forest inventory can provide a huge reduction in human effort required to track forest changes Cost-effective Interesting research Can be performed for an entire forest, not just a few stands.

Assumptions: 

Assumptions Trees are roughly circular or elliptical [Song03] One “highest point”, or treetop Brightest appearance near center [Daley98], [Wang04]

Strategy: 

Strategy Find treetops Attempt to find the boundary of each tree-crown

Treetop Detection: 

Treetop Detection Pass a circular window searching for peaks in the DEM and image intensity Whenever a DEM peak is “near” an intensity peak, mark the midpoint as a treetop (also mark lone intensity peaks)

Tree-Crown Segmentation: 

Tree-Crown Segmentation Initialize balloon as a circle around the treetop Expand until boundary is met (high gradient, low elevation, dark appearance)

Iterations…: 

Iterations…

Automation: 

Automation Process treetops in order from highest to lowest elevation Discard treetops that are inside a tree

Comparison With Hand Segmentations: 

Comparison With Hand Segmentations

Comparison With Hand Segmentations: 

Comparison With Hand Segmentations

Comparison With Hand Segmentations: 

Comparison With Hand Segmentations My algorithm tends to be more generous with the area identified as part of a tree. My algorithm found ~40% more trees because it is more likely to break larger “clumps” into multiple trees than the human segmenter.

Improvements: 

Improvements Reduce the expansion force Increase dependence on intensity gradient Get more hand segmentations Be more conservative with tree shape tests to throw out more segmentations

Future Work: 

Future Work Improve Snake Algorithm Try other methods of segmentation Tree-crown classification by species Change detection (size, death, disease detection, etc.)