Enhanced CNN-LSTM approach for Human Activity Recognition

Abstract

Research in Deep Learning (DL) based computer vision techniques attracted wider attention in the research area. Human activity recognition by digital devices still lacks to perform well in different activity environments. To address this challenge, we proposed the hybrid framework called Enhanced CNN-LSTM approach for the Human Activity Recognition model. In recent days, huge theft and violent activities are happening everywhere, unable to find the culprit even though many technologies exist. The proposed deep learning model monitors and detects human activity through the hybrid approach, which combines both Convolutional Neural Network for spatial and temporal feature extraction and finding the discriminative feature identification using the technique Long Short Term Memory algorithm. The technique LSTM aims to predict the sequences. The experiments are conducted on the UCF10 Human Activity Recognition datasets, which results in improved performance of activity recognition.