Linear algebra holds immense promise of throwing up new surprises in recognizing patterns. The emergence of principal component analysis, which relies heavily on singular value decomposition (SVD), has been one of the drivers of this excitement.

In most of the cases, the presence of a large set of features has often been an obstacle for the efficient analysis of patterns. For analyzing something like hand gestures, the problem compounds further, as data from a third dimension also need to be processed, which is related to time. Is it possible to deploy SVD in such 3D data for recognizing hand gestures? Additionally, since hand gestures involve the movement of multiple fingers, would it be possible to identify the finger whose movement plays a prominent role in such gestures? It is against this backdrop that the authors attempt to propose a gesture recognition method using SVD.

Three-dimensional data captured in connection with the time-dependent motion of each finger is decomposed in each of the m time frames to produce n vectors in each of the three different dimensions. The description of the composition of the final feature set gets blurred here, as the authors do not narrate in precise terms how the final feature set is developed. It is presumed that the (left singular) vectors from the three different dimensions are grouped together and classified as a feature set for a gesture.

The remaining part of this short paper narrates the development of a training set (for developing the model) and a test set (for validating the model) for the different gestures separately for different fingers. The results indicate that gesture recognition is indeed possible and that the tip of the thumb plays a more prominent role than other fingers in analyzing hand gestures.