Homeomorphic Alignment of Edge-Weighted Trees
read more
Citations
Generic initialization for motion capture from 3d shape
Applications of digital topology for real-time markerless motion capture
References
A survey of advances in vision-based human motion capture and analysis
The visual hull concept for silhouette-based image understanding
The Tree-to-Tree Correction Problem
Shock graphs and shape matching
Automatic rigging and animation of 3D characters
Related Papers (5)
Constructing the Tree of Shapes of an Image by Fusion of the Trees of Connected Components of Upper and Lower Level Sets
Frequently Asked Questions (16)
Q2. What have the authors stated for future works in "Homeomorphic alignment of edge-weighted trees" ?
In future works, the authors will take into account more useful information on the model, such as spatial coordinates of data vertices, and include them in their algorithm, for a better robustness.
Q3. What is the way to solve the problem of splitted vertices?
Proposed edit distances, isolated-subtrees distances and top-down distances cannot always match all the model tree, but only subparts, most often unconnected.
Q4. What is the common approach to compare graphs?
An approach widely used to compare two graphs is to search for a sequence of simple primitive operations (called edit operations) that transforms a graph into the other and that has a minimal cost, called the edit distance.
Q5. Why can't a shape be matched to a data tree?
Due to the skeletonization algorithm and to the amount of noise of the shape surface, branchs of skeleton can appear, but without important topological signification.
Q6. How many vertices are there in the tree?
the data tree obtained from the skeleton of the visual hull has a degree bounded by 4, and its number of vertices is between seven and twenty, with a gaussian probability repartition centred on ten.
Q7. What is the way to calculate alignment distance?
Alignment distance is an interesting way in their case for three reasons:it preserves topological relations between trees, it can be computed in polynomial time, and it enables to ”remove edges”, regardless of the rest of the graph, solving the problem of splitted vertices.
Q8. What is the main difficulty of motion capture?
One difficulty of motion capture consists in finding the initial pose of the subject, represented by a 3D shape and constructed using a multiview system.
Q9. What is the undirected graph associated to G?
The undirected graph associated to G is the undirected graph G′ = (V,E), such that {x, y} ∈ E if and only if (x, y) ∈ A or (y, x) ∈ A.
Q10. What is the common problem in the tree?
In the case of a rooted tree, the authors consider that the root rGofG cannot be removed by the cut operation, and then the authors can use the notation Cut(G,K) = Cut(G,K, rG).
Q11. How will the authors use the homeomorphic alignment in future work?
In future works, the authors will take into account more useful information on the model, such as spatial coordinates of data vertices, and include them in their algorithm, for a better robustness.
Q12. How many papers have been published on motion capture?
Motion capture without markers is a highly active research area, as shown by Moeslund and al. [1]:between 2000 and 2006, more than 350 papers on this topic were published.
Q13. What is the weight of the unique path from x to y?
In a weighted tree, the weight of the unique path from x to y, denoted by ω(x, y) is the sum of the weights of all arcs traversed in the path.
Q14. i VP p, j VD?
(4)Proposition 3. Let i ∈ VP \\ {p}, j ∈ VD \\ {d}, ia ∈ anc(i), ja ∈ anc(j),ηcut(∅, ∅) = 0 ηcut(P (i, ia), ∅) = ηcut(F(P, i), ∅) + γ(ω(ia, i), 0) ηcut(F(P, ia), ∅) = ∑i′∈C(ia)ηcut(P (i ′, ia), ∅)ηcut(∅,D(j, ja)) = 0 ηcut(∅,F(D, ja)) = 0 .(5)Proposition 4. Let i ∈ VP \\ {p}, j ∈ VD \\ {d}, ia ∈ anc(i), ja ∈ anc(j).ηcut(P (i, ia),D(j, ja)) =min ηcut(F(P, i), ∅) + γ(ω(ia, i), 0) γ(ω(ia, i), ω(ja, j)) + ηcut(F(P, i),F(D, j)) minjc∈C(j){ηcut(P (i, ia),D(jc, ja))} minic∈C(i){ηcut(P (ic, ia),D(j, ja)) + ∑i′ c ∈C(i)\\icηcut(P (i ′ c, i), ∅)} .(6)Proposition 5. ∀A ⊆ F(P, i), B ⊆ F(D, j),ηcut(A,B) =min minD(j′,j)∈B {ηcut(A,B \\ {D(j ′, j)})} minP (i′,i)∈A {ηcut(A \\ {P (i ′, i)}, B) + ηcut(P (i′, i), ∅)} minP (i′,i)∈A,D(j′,j)∈B {ηcut(A \\ {P (i ′, i)}, B \\ {D(j′, j)})+ηcut(P (i ′, i),D(j′, j))}minP (i′,i)∈A,B′⊆B {ηcut(A \\ {P (i ′, i)}, B \\ B′)+ηcut(F(P, i ′), B′) + γ(Ω(i′), 0)}minA′⊆A,D(j′,j)∈B {ηcut(A \\ A ′, B \\ {D(j′, j)})+ηcut(A ′,F(D, j′)j) + γ(0, Ω(j′))} .
Q15. What is the definition of a tree-matching algorithm?
In the following, after reviewing basic notions, the authors introduce both a new alignment, called homeomorphic alignment, and a robust tree-matching algorithm which may be used for realtime pose estimation.
Q16. What is the main difficulty of motion capture without markers?
After this step, the main difficulty is to match the pattern tree in the data tree, with a good preservation of both topology and distances.