An Adaptive Hybrid Deep Learning Approach for Human Action Recognition
Shreyas Pagare, Rakesh Kumar, Sanjeev Kumar Gupta · Advances in intelligent systems research/Advances in Intelligent Systems Research · 2025
Human Activity Recognition (HAR) technology, which is focused on identifying and analyzing human activities, has gained significant interest in recent years.Traditional approaches have employed manually designed features to identify human activities, leading to limited feature extraction.Neural network detectors are increasingly used in personal and portable devices to detect and recognize human actions.Nevertheless, unimodal methods rely on a solitary sensing modality and employ machine learning techniques to identify human activities.A deep learning-based Human Activity Recognition (HAR) model called Adaptive Hybrid Deep Attentive Network (AHDAN) will be created to address these abstract concepts.This model will combine a 3D Convolutional Neural Network (1DCNN) with gated Recurrent Units (GRU) to enhance the recognition process.Additionally, the parameters of the network will be optimized to improve the recognition process further.Through comprehensive experimental assessments on the UCF101 benchmark dataset, we have established that our proposed method surpasses existing state-of-the-art techniques in action recognition.These findings underscore the capability of our approach to enhance future research in video action recognition.This study presents a novel method for identifying actions in video content.The technique combines attention-based mechanisms with a long short-term memory network and an improved, optimized 3D Convolutional Neural Network to achieve effective action recognition.