Spatiotemporal Action Recognition in Videos Using ConvLSTMwith Attention: A Comparative Analysis and Implementation
Kohari, Mohammed Kaif · Zenodo (CERN European Organization for Nuclear Research) · 2025
This dissertation explores the application of Convolutional Long Short-Term Memory (ConvLSTM) networks for video action recognition. ConvLSTM integrates the spatial feature extraction capabilities of CNNs with the temporal modeling strengths of LSTM networks, enabling the model to capture both spatial and temporal dependencies within video data. The primary goal of this research is to develop an efficient, scalable model capable of recognizing actions in real-time. The UCF-101 dataset is used to evaluate the effectiveness of the proposed model, with performance compared against traditional CNN and LSTM approaches. Additionally, various preprocessing techniques and hyperparameter configurations are examined to understand their impact on model performance.