A Robust Model of Human Activity Recognition using Independent Component Analysis and XGBoost

In computer vision and pattern recognition, image-based human activity detection has gained popularity as a research topic. The process of recognizing human behavior from an image is called activity recognition. Here, we put an XGBoost into practice to create an activity recognition system. The suggested method applies an HSV colour transformation in its initial phases to improve video frame clarity. Subsequently, it utilizes filters to minimize noise. Multiple object tracking (MOT) and VIBE techniques extract the silhouette. For feature extraction, we employ both Textone maps and FAST. In the next step, In order to conduct feature discrimination, Independent Component Analysis (ICA) is used to identify the most informative independent components that capture the data’s underlying structure. Subsequently, the features are inputted into XGBoost and categorized into pertinent human activities according to their final features. The experimental procedure makes use of the SBUInteraction dataset to attain a 91% recognition rate.