What is this project?
This project is an undergraduate capstone project focusing on detecting violent behavior in videos using deep learning techniques.
The goal of the system is to classify short video clips as either violent or non-violent by learning spatiotemporal features from video data.
What I worked on
- Training and evaluating a 3D CNN model for video-based violence detection
- Preparing and preprocessing video datasets for model input
- Analyzing model performance and limitations
- Implementing the frontend interface to visualize inference results
Tech Snapshot
3D CNN · ViViT · Video Classification · Deep Learning · Computer Vision
Reflection
This project deepened my understanding of video-based deep learning models and the challenges of learning spatiotemporal features from real-world data.
It also highlighted the trade-offs between model complexity, training cost, and performance, and strengthened my interest in applying AI techniques to real-world safety-related problems.