Abstract | ||
---|---|---|
Optical motion capture system is widely used to acquire human motions by capturing the trajectories of markers that are attached to the body. Identifying the marker trajectories is challenging but indispensable in most of real applications. Conventional methods rely on either labor-intensive manually labeling or auto-labeling with assumption of pose similarity to the topological model. This paper presents a novel method to flexibly label markers from human motion capture sequences. The point sets in a rigid segment defined in the topological model are firstly clustered by using the spectral clustering algorithm. For each rigid segment, a local k-d tree is constructed to robustly match two point sets without pose similarity assumption. To match all rigid bodies with those in topological model for efficiently and correctly labeling, the labeling process is carefully designed using the articulated structure of acquired data. Experiments show that our method outperforms conventional methods in accuracy and is robust when labeling markers in motion capture sequences from different subjects. |
Year | DOI | Venue |
---|---|---|
2011 | null | BIODEVICES 2011 |
Keywords | Field | DocType |
Optical motion capture,Label markers,Local k-d tree,Clinical gait analysis | k-d tree,Electronic engineering,Theoretical computer science,Engineering | Conference |
Volume | Issue | Citations |
null | null | 0 |
PageRank | References | Authors |
0.34 | 0 | 2 |
Name | Order | Citations | PageRank |
---|---|---|---|
Wu Huang | 1 | 0 | 0.34 |
Shihong Xia | 2 | 287 | 26.18 |