Saturday, September 26, 2015

Reading 8 : Rubine's Features

Citation:
Rubine, Dean. Specifying gestures by example. Vol. 25. No. 4. ACM, 1991.
Publication Link

Summary:
Rubine's feautres are often considered the pioneering work in gesture recognition. This paper explains the intuition behind these gestures, and the implementation of a toolkit called GRANDMA. The paper details the MVC architecture used in the toolkit, the statistic features of a gesture, how a classifier is training and also eager and multi-finger gestures.

Discussion:
The main takeaway from this paper are the eleven features (which will be discussed in the next reading) , and the approach to train a linear classifier using these features.

The main aim is to find the class that maximizes the value of v according to the following relation:


The computation of the weights for each feature of each class is done with the following relations:


The paper also proceeds to define a probabilistic heuristic for rejection and also describes the Mahalonobis distance which measures how many standard deviations away a gesture is from a given class.





No comments:

Post a Comment