Related Experiment Video
Updated: Jun 27, 2026
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
An efficient hybrid CNN-transformer framework for real-time weapon detection and face recognition.
1School of Computer Science and Engineering, Vellore Institute of Technology, Chennai, Tamil Nadu, India.
Frontiers in Artificial Intelligence
|June 26, 2026
Summary
This study introduces ConViDeTR, a hybrid deep learning framework for smart surveillance. It achieves high accuracy in real-time weapon detection and face recognition, outperforming existing methods.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Deep Learning
Background:
- Smart surveillance demands accurate, real-time weapon detection and face recognition.
- Existing Convolutional Neural Network (CNN) or Vision Transformer (ViT) methods struggle with feature extraction and complex conditions.
- Robustness against occlusion, illumination changes, and complex backgrounds is crucial.
Purpose of the Study:
- To present ConViDeTR, a novel hybrid deep learning framework.
- To enable synchronous weapon detection and face recognition within a unified system.
- To enhance accuracy and robustness in intelligent surveillance.
Main Methods:
- Integration of CNN, Vision Transformer (ViT), and Detection Transformer (DETR) architectures.
- Introduction of a deep feature fusion layer for integrating diverse feature types.
- Development of a shared feature space for synchronous task execution.
Main Results:
- Achieved 98.9% accuracy in weapon detection and 97.34% in face recognition.
- Demonstrated superior performance compared to existing techniques on benchmark datasets.
- Real-time processing capability with 25-30 FPS and low latency.
Conclusions:
- ConViDeTR offers an effective, robust, and scalable solution for intelligent surveillance.
- The hybrid approach overcomes limitations of standalone CNN and ViT models.
- The framework supports next-generation smart surveillance systems.