Parametrized SOMs for hand posture reconstruction

Fingertip detection with morphology and geometric calculation

“…If there are good detectors which can detect hand features such as fingers, then hand posture can be better represented by hand structure models. Studies modeling the human hand for gesture recognition are cited in [9], [10]. These studies try to reconstruct the hand model from detected fingers using the inverted kinetic model.…”

“…In some finger detection studies such as [9], [10], it is assumed that the hand has already been detected and thus a hand detection step is not needed. This assumption does not hold in the general case since the hand can be confused with other objects having the same color.…”

“…In [9], [10], the authors use Gabor filters to extract features and detect fingertips by using a local linear mapping (LLM) network. They then compute parameters of hand structure using parameterized self-organizing map (PSOM) to reconstruct the 3D hand shape.…”

“…The key issue in these techniques is the definition of the hand structure model. In some complex systems such as GREFIT [9], [10], the hand model is defined by links and joints. This approach uses fingertips, which are usually detected by applying a learning model, to reconstruct the hand model.…”

Estimating 3D hand pose from a cluttered image

“…Hand pose is estimated based on the geometric moments of the input hand image. In [15] 3D locations of fingers are estimated from a stereo image, and they are used to infer 3D joint angles. In [18] hand pose is estimated from a single image using shadow information and assuming a calibrated light source.…”