CNN+LSTM: Automatic Action Recognition

Project Scope

This project leverages deep learning to advance action recognition in video data, combining Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks for robust spatial and temporal feature analysis. It showcases the potential for automated activity classification, with applications in robotics, surveillance, and smart systems.

Utilizing the UCF datasets (UCF11, UCF50, UCF101) as benchmarks, the project achieved notable accuracy rates across diverse action categories by training the models on the PSC Bridges-2 supercomputer cluster, utilizing Tesla V100-SXM2-32GB GPUs. With a focus on preprocessing, model optimization, and real-world video testing, this project highlights the growing role of machine learning in activity recognition and intelligent automation. Future work will explore integrating this system into additive manufacturing workflows and robotic control systems for advanced monitoring and optimization.

Highlights

• CNN-LSTM model for spatial and temporal action recognition

• Classification with UCF datasets (UCF11, UCF50, UCF101)

• Deploying trained model  with custom video action recognition setup

• Future potential for additive manufacturing process monitoring-control, robotics, etc.