Effectiveness of Classifiers to Identify Hand Gestures with Motion Capture Coordinate Markers

In a world where technology is increasingly moving towards less wires, less devices, and more portability, the ability of a system to accurately interpret hand gestures is extremely valuable. The innovation of small, inexpensive cameras only makes this even more practical. The practicality of such a system rests on the assumption that hand gestures can be recognized from participants based solely on the shape of their hand. The dataset used for this research, while captured with node markers on a special glove, is formatted independent of a marker indexing system and could be applied in conjunction with a marker identification system for real world application of hand gesture classification. This paper presents an analysis and comparison of the effectiveness classifiers have when trying to determine hand gestures using motion capture coordinate markers provided by a Viacon camera. Stochastic Gradient Descent, Decision Trees, Logistic Regression, Random Forests, and Bagging are applied in this study. Overall, the Random Forests classifier performed the best with respect to performance measures such as recall, F
₁
score, and precision.