A human activity recognition framework using max-min features and key poses with differential evolution random forests classifier

This paper presents a novel framework for human daily activity recognition that is intended to rely on few training examples evidencing fast training times, making it suitable for real-time applications. The proposed framework starts with a feature extraction stage, where the division of each activity into actions of variable-size, based on key poses, is performed. Each action window is delimited by two consecutive and automatically identified key poses, where static (i.e. geometrical) and max-min dynamic (i.e. temporal) features are extracted. These features are first used to train a random forest (RF) classifier which was tested using the CAD-60 dataset, obtaining relevant overall average results. Then in a second stage, an extension of the RF is proposed, where the differential evolution meta-heuristic algorithm is used, as splitting node methodology. The main advantage of its inclusion is the fact that the differential evolution random forest has no thresholds to tune, but rather a few adjustable parameters with well-defined behavior.