Exploring Activity Transitions in the STA-based Activity Recognition

The STA (Scanpath Trend Analysis) based activity recognition approach considers both binary sensors and activity transitions to predict activity for a given instance. It is advantageous over other approaches as it provides higher accuracy with less computational complexity. This study explores the effects of activity transitions on the STA-based activity recognition approach. It investigates how the accuracy of the STA-based approach is affected when activity transitions with lower probabilities are considered, and when a different approach is used to compute transition probabilities. The experiments with the dataset previously used for the evaluation of the approach does not reveal considerable differences in the accuracy.