← Back to Publications List

Enhanced Human Detection in Challenging Environments Using Visible and Infrared Thermal Imaging

Students & Supervisors

Student Authors
Md. Maidul Islam Shehab
Bachelor of Science in Electrical & Electronic Engineering, FE
Debashish Kumar Ghosh
Bachelor of Science in Electrical & Electronic Engineering, FE
Shajib Hossen
Bachelor of Science in Electrical & Electronic Engineering, FE
Md. Rafsan Jany
Bachelor of Science in Electrical & Electronic Engineering, FACULTY OF ENGINEERING
Supervisors
Abu Shufian
Lecturer, Faculty, FE

Abstract

This study presents an intelligent human detection system designed for low-visibility environments, including smoke-filled rooms, darkness, or low-light conditions. Leveraging the capabilities of visible and thermal infrared imaging, the system integrates a 64×64-pixel Adafruit thermal camera module and a Raspberry Pi 4 microcontroller with the lightweight Tiny YOLOv4 object detection algorithm. Thermal image data is upscaled to 128×128 resolution using bilinear interpolation to facilitate deep learning-based classification. A custom dataset of 1,990 infrared images was compiled from ten human subjects performing four everyday actions in varying levels of smoke and lighting. The model achieved an 87% accuracy rate with an average processing speed of 8.8 milliseconds per frame. Detailed metrics confirm the system's robustness, including precision, recall, and F1-score. Furthermore, it supports real-time alert functionality via GSM modules for emergency response. Despite resolution and dataset size limitations, the proposed solution highlights the feasibility of deploying affordable, low-power human detection systems in critical safety and surveillance scenarios.

Keywords

Human detection Image processing Raspberry Pi Infrared Thermal camera CNN Tiny Yolo v4.

Publication Details

  • Type of Publication: Conference 
  • Conference Name: IEEE Region 10 Symposium 2025 (TENSYMP2025)
  • Date of Conference: 07/07/2025 - 07/07/2025
  • Venue: University of Canterbury, Christchurch, New Zealand
  • Organizer: IEEE Region 10
View on Publisher