SAE – Computer Vision
Computer vision system to detect human silhouettes and verify safety equipment wear.
- Context
- Coursework · BUT R&T
- Field
- Data & vision · Security
- Tools & technologies

01
Project Context
This project focuses on computer vision for human silhouette detection and security alert management. The main goal was not full facial recognition, but to detect a silhouette’s presence and check whether it wears a yellow safety vest (authorized staff). If a silhouette without a yellow vest is detected, an intrusion is reported.
02
Methods & Components
- Python language with OpenCV library for image processing and detection
- Silhouette detection: Using HOG (Histogram of Oriented Gradients) descriptor combined with filtering and morphology
- Background subtraction: MOG2 algorithm to isolate foreground movement and objects
- Color detection: Using HSV color space to spot yellow vests within the silhouette area
- State management: Application logic to track states (silhouette detected → vest present → no alert / otherwise alert)
- Remote logging: Upon intrusion, sending a log to a remote virtual machine acting as a log server
03
Skills Acquired
- OpenCV Proficiency
- Image processing pipeline in Python
- Background Subtraction
- Implementation and tuning of MOG2 for motion detection
- HOG Descriptors
- Using filtering and morphological techniques to robustify detection
- HSV Color Detection
- Simple classification to identify safety equipment
- State Management
- Designing logic to decide when to raise an alert
- Remote Logging
- Network communication and log format to a remote VM