Authors: Murad Hasan, Shahriar Iqbal, Md. Billal Hossain Faisal, Md. Musnad Hossin Neloy, Md. Tonmoy Kabir, Md. Tanzim Reza, Md. Golam Rabiul Alam, Md Zia Uddin

Published on: February 05, 2024

Impact Score: 8.35

Arxiv code: Arxiv:2402.03417

Summary

• What is new: A novel deep learning-based hybrid fusion model for detecting stalking from videos, significantly improving accuracy.
• Why this is important: Physical stalking detection in public spaces is under-researched, despite its prevalence and the potential to prevent crimes.
• What the research proposes: A deep learning model that analyzes video frames using facial landmarks, head pose, relative distance, and spatio-temporal features to classify stalking behavior.
• Results: The model achieved 89.58% accuracy in detecting stalking incidents, outperforming existing methods.

Technical Details

Technological frameworks used: Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM), Multilayer Perceptron (MLP)

Models used: Hybrid fusion model combining numerical data analysis and spatio-temporal video frame analysis

Data used: Dataset of stalking and non-stalking videos from movies and TV series

Potential Impact

Security and surveillance companies, public safety departments, social media platforms with live video streaming features

Want to implement this idea in a business?

We have generated a startup concept here: SafeGuard AI.