Reading Assignment: Gestures without Libraries, Toolkits, or Training: A $1 Recognizer for User Interface Prototypes

Reference Information
Title: Gestures without Libraries, Toolkits, or Training: A $1 Recognizer for User Interface Prototypes
Authors: Jacob O. Wobbrock, Andrew D. Wilson, Yang Li
Citation: "Gestures without Libraries, Toolkits or Training: A $1 Recognizer for User Interface Prototypes", Jacob O. Wobbrock, Andrew D. Wilson, Yang Li, Proceedings of the 20th annual ACM symposium on User interface software and technology, pp. 159-168, 2007.

Summary
This paper discussed the $1 recognizer, a gesture recognition algorithm designed to be cheap, simple, and easy-to-use such that even novice programmers can include it in their own interface systems. Many gesture recognition algorithms rely on complicated math (such as that presented in the Rubine paper) or computationally expensive methods that limit the number of programmers that can implement such a system.

When given a gesture as input, the the $1 recognizer works through a series of four steps to recognize the given gesture. The first step resamples the path of the gesture to produce a path with N equally-spaced points. Step two rotates the gesture based on an angle found using a seed and search approach. Step three scales the gesture non-uniformly, then translates it to a particular reference point. Finally, step four conducts the recognition by comparing the modified gesture to a set of stored templates. Limitations of this recognizer include requiring comparison to templates, being rotation, scale, and position invariant, and being unaware of time as related to the gestures.

The $1 recognizer was tested against two other algorithms, the Rubine classifier and Dynamic Time Warping (DTW). Tests were conducted by having users provide a series of gestures at varying speeds, then using the three algorithms to recognize the gestures. It was determined that medium speed gestures are recognized more accurately than slow or fast gestures, most likely due to the balance between speed and accuracy. It was also determined that the $1 recognizer had accuracies similar to DTW and better than Rubine for the experiment that was conducted.

Thoughts
I think that it's a great idea to provide a recognition algorithm that is simple, easy-to-understand, and easy for novice programmers to include in their own work. This could serve to not only increase awareness of gesture recognition, but also to increase the number and range of ideas surrounding gesture recognition by having systems implemented by a much wider range of programmers with differing backgrounds.

This paper made for a very good read, and it was easy to understand the workings of the algorithm. It made many references to the Rubine paper that was also assigned as reading for this course, so having previously read that paper made it much easier to understand some of the motivations and the structure of the recognizer mentioned here. In addition, having been written in 2007, this paper is much more current than the others that we have read so far. While it is nice to see the foundations of sketch recognition, it was also nice to read of some more current technology and how it actually applied the foundations that we have learned.

One problem that I have with the paper is that it repeatedly mentioned a major goal of the recognizer being that it should be easily usable by novice programmers; however, it is later mentioned that the programming ease has yet to be tested. Therefore, it is unknown whether or not the algorithm actually accomplishes this major goal that was set for it. However, the evaluation that was provided was very helpful in seeing how this recognizer compares with others.