ik oct 24 2000

Slide1: 

Tree data from remote imageries Marv 14 Tue 24. Oct. 2000 Ilkka Korpela

Slide2: 

Tree data Location attributes: {X, Y, Z |(top),(base)} Object description: {species, dx, h, health,..} Remote imageries POS, Active: spectral & radiometric resolution

Slide3: 

Location & object description  3-D reconstruction of trees POS ALS (active sensing)

Slide5: 

Aerial photography Model of central projection: * 6 + 3 parameters for exterior & inner orientation, * G: E3 E2 is an onto mapping (surjektio) G:(X,Y,Z)-> (x,y) * G-1 is an injection. => reconstruction is ill-posed

Slide6: 

3-D reconstruction of tree tops (apexes, crown tops) 1. Find tree tops from all projections 2. Solve the correspondence problem for conjugate entities 3. Calculate 3-d coordinates 4. Verify matching

Slide7: 

Finding tree tops from single views 1/5 “Image analysis is a process of discovering, identifying and understanding patterns that are relevant to the performance of an image based task” (Gonzalez & Woods 1993, p. 571).  Model database (modelbase)  Feature detector  Hypothesizer  Hypothesis verifier SYSTEM

Slide8: 

Finding tree tops from single views 2/5  Scene constancy  Image-model spaces  Number of objects in the model database  Number of objects in an image and possibility of occlusion Consider the task:

Slide9: 

Finding tree tops from single views 3/5

Slide10: 

Finding tree tops from single views 4/5 * Segmentation based methods * Model-based methods - Template matching (synthetic) g

Slide11: 

Finding tree tops from single views 5/5

Slide12: 

Matching and reconstruction 1/6 ILL-posed task => restrict & condition  increase views n   , p(solution exist)  1  restrict the search space, a tree top can not be just anywhere EPIPOLAR CONSTRAINT

Slide13: 

Matching and reconstruction 2/6

Slide14: 

Matching and reconstruction 3/6

Slide15: 

Matching and reconstruction 4/6

Slide16: 

Matching and reconstruction 5/6 Problems: + scale invariancy =>enlarge modelbase, increase operator input + different reflection properties of tree species => synthetic matching + computational complexity O(n4) or O(n3) => decrease n, multiresolution approach, calculate T-matching in Frequancy-domain + Terrain model needed, preferably high accuracy, user input needed in limiting search space in Z.

Slide17: 

Matching and reconstruction 6/6

Slide18: 

MOTIVATION