Traditional security checkpoints (e.g., airports, hotels, event venues) rely heavily on manual ID verification and human surveillance, which are time-consuming and prone to human error, vulnerable to identity fraud, forged documents, and look-alikes, inefficient in detecting wanted, banned, or high-risk individuals in real-time. Given the increasing need for automated, intelligent surveillance systems, there is a strong demand for a non-intrusive, reliable, and scalable solution to screen individuals based on biometric identity especially facial recognition without interrupting regular flow.
A real-time security screening system, using face recognition with OpenCV. The system aims to scan individuals via a webcam at check- points (e.g., hotels, airports). If a person’s face matches an entry in the system’s dataset of known threats, runaway criminals or wanted individuals, the system triggers a threat alarm and sends an email alert to authorities. Otherwise, a safe alarm is triggered, allowing them to proceed.
As of July 2025, this project is a part of GirlScript Summer of Code 2025. Please read the Readme File carefully for understanding the project workflow. For contribution tips and extended documentation, see the Learn Guide. Only the issue with the label gssoc25 are open for gssoc contributors right now.
-
The system scans individuals and detects faces.
-
Compares detected faces against a dataset of wanted/banned/runaway individuals.
-
The system takes 10 seconds to analyze and confirm the match before coming to a conclusion.
-
If no match is found a safe_alarm is triggered indicating the person is harmless and safe to go.
-
If a match is found a threat_alarm is triggered and an email is sent directly to the authorities notifying them of the potential threat.
-
if a match is found, the individual's image is automatically logged in the system for future verification and other legal procedures.
-
Advanced data augmentation: Everytime an image is added, run the data augmentation.py script to add more diverse set images for each person.
-
Dataset: The dataset contains folders for each wanted individual. The folder contains images of that individual. Larger the no. of images, better the precision of the system. The system has been tested with personal images. For now, sample stock images are used to fill the folder.
-
Face Encoding & Labeling: The system loads all images, extracts facial features using the face_recognition model , and encodes them into numerical vectors.It then assigns each encoding to the person’s name based on the folder structure.
-
Real time face recognition: Scans the individuals face using OpenCV. Detects and extracts faces in real time. Compares detected faces against stored encodings using distance metrics. If similarity is above a certain threshold (< 0.4), the person is identified. The threshold is intentionally set lower in order to avoid false positives. While false negatives can be handled with additional procedures, false positives can cause serious harassment and authority disturbance issues and are therefore minimized.
-
Identity Confirmation: The system takes 10 seconds to analyze and confirm the match. If the match remains consistent for 10 seconds and the alarm has not been trigerred for last 30 seconds, it triggers the alarm based on the match type: safe_alarm for "No match" and threat_alarm for "match".
-
Email Notification: If a threat is identified and threat_alarm is trigerred an email is sent to authorities including the name, photo, time and Ip location of the individual. The authorities can be the organisation security, admin, police authorities, local authorities, public safety etc.
-
SMS Notification: If a threat is identified and threat_alarm is triggered, an SMS alert is sent to the concerned authorities. The message includes key details such as the individual's name, time of detection and Ip location of the individual.
-
Logging Matched faces: if a match is found, the individual's image is automatically logged in the system for future verification and other legal procedures.
-
Advanced data augmentation: Everytime an image is added, run the data augmentation.py script to add more diverse set images for each person.
- This is a functional prototype designed for a security checkpoint use case. It demonstrates the core facial recognition and threat-detection logic of a real-time screening system. The project is modular and can be extended to integrate with IoT hardware or GUI interfaces as needed.
- Due to a deliberately lower matching threshold to reduce false positives, the system might produce false negatives in few cases.
- To improve accuracy and reliability, it is recommended to use a larger number of diverse images per individual in the dataset.
This project is licensed under the MIT License.
