AI Based Human Activity Recognition

Fathima Sherief, Jogimol Joseph

Abstract: This project aims to develop a Human Activity Recognition (HAR) system using deep learning, combining CNNs for spatial feature extraction and LSTMs for temporal analysis. The system processes video data to classify activities as "Normal" or "Suspicious", with real - time results displayed through a Tkinter - based GUI. It uses datasets like NTU RGB+D and custom videos for training and testing. Despite challenges like varying accuracy and long training times, the system offers a reliable tool for real - time surveillance and security applications.

Keywords: Deep learning, Convolutional neural network, LSTM, GUI