Abstract
The paper presents an approach for ensemble-based feature selection in human activity recognition. The goal is to select an important discriminating features to recognize the human activities in videos and removing the irrelevant redundant features. The features are extracted based on spatiotemporal orientation energy and template matching. Due to robust and accurate ensemble models with low variability and biases, Gradient Boosting and Random Forest are applied to identify the relevant features. Support Vector Machine with linear kernel is used to classify the activities. The experiments have tested on KTH dataset. The results show an improvement in accuracy (better by 1.51%) and the features are reduced by 99.2%. The Comparisons to related works were given.