Data Trimming Methods to Improve Gesture Classification

This paper introduces a data processing method to enhance the performance of a gesture classification model. When tested on the UTD-MHAD dataset, the HMM model initially rendered a poor performance due to seemingly resembling gestures. To tackle this problem, data has been altered via normalization and selection of significant joints that determine the gesture. Refining data prior to classifying generates a better performance in both HMM and LSTM models, highlighting the significance of data processing across different types of classification models.