An Adaptive Deep Learning Framework for Human Activity Recognition using Video Data
Chejarla Raja S, D. Arivazhagan, S Lavanya, C Rajeswari, Banu S, K Rahmaan · 2025
This research presents a dual-pathway SlowFast network in order to provide a novel approach for the recognition of human actions. The method has the ability to capture low-intensity as well as high-intensity activities efficiently. The model is trained on two different datasets: one on the routine activities, such as sitting and walking, and the other on complex sports and leisure activities. The SlowFast network increases the accuracy in a wide variety of scenarios by incorporating high-level spatial context along with fast motion characteristics using a dual-path architecture. The proposed model outperforms the conventional HAR methods on different grounds such as adaptability and accuracy, and thus aptly useful for applications such as security surveillance, healthcare, and sports analysis. This approach brings forth a robust system able to identify a range of human actions in static as well as dynamic environments by removing some of the limitations from existing HAR models.