Pattern Recognition and Classification on multidimensional vector time-series
Gesture Recognition, Autoregressive Models, Multidimensional Vector Time Series
The present work highlights the recognition of static human gestures, employing autoregressive models and the analysis of multidimensional vector time series. Using the MediaPipe library to capture gestures through a webcam, 140 videos were recorded and analyzed, providing a dataset for analysis. The third-order autoregressive model proved to be an interesting methodology for identifying specific patterns and characteristics of gestures. The analysis of the model’s coefficients for each landmark and coordinate revealed distinct aspects of each movement, emphasizing the complexity and variability inherent in human gestures. The three-dimensional visualization of the coefficients and the color differentiation facilitated the comparison between gestures and the identification of trends and variations. Furthermore, the statistical analysis of the coefficients through boxplots provided a clearer understanding of the stability and variation of gestures.